Update goleveldb

Directory auto-complete.

.gitignore

@@ -4,7 +4,9 @@ syncthing.exe
 *.zip
 *.asc
 *.sublime*
-discosrv
 .jshintrc
 coverage.out
 files/pidx
+bin
+perfstats*.csv
+coverage.xml

.travis.yml

@@ -1,20 +0,0 @@
-language: go
-
-go:
-   - tip
-
-install:
-   - export PATH=$PATH:$HOME/gopath/bin
-   - ./build.sh setup
-   - go get code.google.com/p/go.tools/cmd/cover
-   - go get github.com/mattn/goveralls
-
-script:
-   - ./build.sh test-cov
-
-after_success:
-   - goveralls -coverprofile=coverage.out -service=travis-ci -package=calmh/syncthing -repotoken="$COVERALS_TOKEN"
-
-env:
-   global:
-      secure: "zEV2h2XtKHNLVdXJjM4LA/VjMfLVydm6goF+ARit+nOSGxGoH7f7jIdzJzhxgh7shKG93q61eLO1Tug+WBMYB2EpBuYnTB5AIMYhCDwNI8C4uBV6c3brHfcrie7MASNao8TID2QScASKNFFWvjv/i1Ccn5ztxdcQuhSsNjGZp8A="

AUTHORS

@@ -0,0 +1,29 @@
+# This is the official list of Syncthing authors for copyright purposes.
+
+Aaron Bieber <qbit@deftly.net>
+Alexander Graf <register-github@alex-graf.de>
+Andrew Dunham <andrew@du.nham.ca>
+Audrius Butkevicius <audrius.butkevicius@gmail.com>
+Arthur Axel fREW Schmidt <frew@afoolishmanifesto.com> <frioux@gmail.com>
+Ben Sidhom <bsidhom@gmail.com>
+Brandon Philips <brandon@ifup.org>
+Caleb Callaway <enlightened.despot@gmail.com>
+Chris Joel <chris@scriptolo.gy>
+Daniel Martí <mvdan@mvdan.cc>
+Emil Hessman <emil@hessman.se>
+Felix Ableitner <me@nutomic.com>
+Felix Unterpaintner <bigbear2nd@gmail.com>
+Gilli Sigurdsson <gilli@vx.is>
+Jakob Borg <jakob@nym.se>
+James Patterson <jamespatterson@operamail.com> <jpjp@users.noreply.github.com>
+Jens Diemer <github.com@jensdiemer.de> <git@jensdiemer.de>
+Jochen Voss <voss@seehuhn.de>
+Lode Hoste <zillode@zillode.be>
+Marcin Dziadus <dziadus.marcin@gmail.com>
+Michael Tilli <pyfisch@gmail.com>
+Philippe Schommers <philippe@schommers.be>
+Phill Luby <phill.luby@newredo.com>
+Ryan Sullivan <kayoticsully@gmail.com>
+Tully Robinson <tully@tojr.org>
+Veeti Paananen <veeti.paananen@rojekti.fi>
+Vil Brekin <vilbrekin@gmail.com>

CONTRIBUTING.md

@@ -1,14 +1,73 @@
-Please do contribute! If you want to contribute but are unsure where to
-start, the [Contributions Needed
-topic](http://discourse.syncthing.net/t/contributions-needed/49)
-lists areas in need of attention.
+## Reporting Bugs
+
+Please file bugs in the [Github Issue
+Tracker](https://github.com/syncthing/syncthing/issues). Include at
+least the following:
+
+ - What happened
+
+ - What did you expect to happen instead of what *did* happen, if it's
+   not crazy obvious
+
+ - What operating system, operating system version and version of
+   Syncthing you are running
+
+ - The same for other connected devices, where relevant
+
+ - Screenshot if the issue concerns something visible in the GUI
+
+ - Console log entries, where possible and relevant
+
+If you're not sure whether something is relevant, erring on the side of
+too much information will never get you yelled at. :)
+
+## Contributing Translations
+
+All translations are done via
+[Transifex](https://www.transifex.com/projects/p/syncthing/). If you
+wish to contribute to a translation, just head over there and sign up.
+Before every release, the language resources are updated from the
+latest info on Transifex.
+
+## Contributing Code
+
+Every contribution is welcome. If you want to contribute but are unsure
+where to start, any open issues are fair game! Be prepared for a
+[certain amount of review](https://discourse.syncthing.net/t/733); it's
+all in the name of quality. :) Following the points below will make this
+a smoother process.
+
+## Coding Style
+
+- Follow the conventions laid out in [Effective Go](https://golang.org/doc/effective_go.html)
+  as much as makes sense.
+
+- All text files use Unix line endings.
+
+- Each commit should be `go fmt` clean.
+
+- The commit message subject should be a single short sentence
+  describing the change, starting with a capital letter.
+
+- Commits that resolve an existing issue must include the issue number
+  as `(fixes #123)` at the end of the commit message subject.
+
+- Imports are grouped per `goimports` standard; that is, standard
+  library first, then third party libraries after a blank line.
+
+- A contribution solving a single issue or introducing a single new
+  feature should probably be a single commit based on the current
+  `master` branch. You may be asked to "rebase" or "squash" your pull
+  request to make sure this is the case, especially if there have been
+  amendments during review. 
 
 ## Licensing
 
-All contributions are made under the same MIT License as the rest of the
-project, except documentation which is licensed under the Creative
-Commons Attribution 4.0 International License. You retain the copyright
-to code you have written.
+All contributions are made under the same GPL license as the rest of the
+project, except documentation, user interface text and translation
+strings which are licensed under the Creative Commons Attribution 4.0
+International License. You retain the copyright to code you have
+written.
 
 When accepting your first contribution, the maintainer of the project
 will ensure that you are added to the CONTRIBUTORS file. You are welcome
@@ -16,8 +75,8 @@ to add yourself as a separate commit in your first pull request.
 
 ## Building
 
-[See the
-documentation](http://discourse.syncthing.net/t/building-syncthing/44)
+[See the documentation](http://discourse.syncthing.net/t/44) on how to
+get started with a build environment.
 
 ## Branches
 
@@ -44,7 +103,9 @@ Yes please!
 
 ## Style
 
-`go fmt`
+ - `go fmt`
+
+ - Unix line breaks
 
 ## Documentation
 
@@ -52,5 +113,4 @@ Yes please!
 
 ## License
 
-MIT
-
+GPLv3

CONTRIBUTORS

@@ -1,10 +0,0 @@
-Aaron Bieber <qbit@deftly.net>
-Andrew Dunham <andrew@du.nham.ca>
-Arthur Axel fREW Schmidt <frew@afoolishmanifesto.com>
-Ben Sidhom <bsidhom@gmail.com>
-Brandon Philips <brandon@ifup.org>
-James Patterson <jamespatterson@operamail.com>
-Jens Diemer <github.com@jensdiemer.de>
-Philippe Schommers <philippe@schommers.be>
-Ryan Sullivan <kayoticsully@gmail.com>
-Veeti Paananen <veeti.paananen@rojekti.fi>

Godeps/Godeps.json

@@ -1,45 +1,49 @@
 {
-	"ImportPath": "github.com/calmh/syncthing",
-	"GoVersion": "go1.3",
+	"ImportPath": "github.com/syncthing/syncthing",
+	"GoVersion": "go1.3.3",
 	"Packages": [
-		"./cmd/syncthing",
-		"./cmd/assets",
-		"./discover/cmd/discosrv"
+		"./cmd/..."
 	],
 	"Deps": [
-		{
-			"ImportPath": "bitbucket.org/kardianos/osext",
-			"Comment": "null-13",
-			"Rev": "5d3ddcf53a508cc2f7404eaebf546ef2cb5cdb6e"
-		},
 		{
 			"ImportPath": "code.google.com/p/go.crypto/bcrypt",
-			"Comment": "null-212",
-			"Rev": "1064b89a6fb591df0dd65422295b8498916b092f"
+			"Comment": "null-216",
+			"Rev": "41cd4647fccc72b0b79ef1bd1fe6735e718257cd"
 		},
 		{
 			"ImportPath": "code.google.com/p/go.crypto/blowfish",
-			"Comment": "null-212",
-			"Rev": "1064b89a6fb591df0dd65422295b8498916b092f"
+			"Comment": "null-216",
+			"Rev": "41cd4647fccc72b0b79ef1bd1fe6735e718257cd"
 		},
 		{
 			"ImportPath": "code.google.com/p/go.text/transform",
-			"Comment": "null-87",
-			"Rev": "c59e4f2f93824f81213799e64c3eead7be24660a"
+			"Comment": "null-90",
+			"Rev": "d65bffbc88a153d23a6d2a864531e6e7c2cde59b"
 		},
 		{
 			"ImportPath": "code.google.com/p/go.text/unicode/norm",
-			"Comment": "null-87",
-			"Rev": "c59e4f2f93824f81213799e64c3eead7be24660a"
+			"Comment": "null-90",
+			"Rev": "d65bffbc88a153d23a6d2a864531e6e7c2cde59b"
 		},
 		{
-			"ImportPath": "code.google.com/p/snappy-go/snappy",
-			"Comment": "null-15",
-			"Rev": "12e4b4183793ac4b061921e7980845e750679fd0"
+			"ImportPath": "github.com/AudriusButkevicius/lfu-go",
+			"Rev": "164bcecceb92fd6037f4d18a8d97b495ec6ef669"
 		},
 		{
-			"ImportPath": "github.com/golang/groupcache/lru",
-			"Rev": "a531d51b7f9f3dd13c1c2b50d42d739b70442dbb"
+			"ImportPath": "github.com/bkaradzic/go-lz4",
+			"Rev": "93a831dcee242be64a9cc9803dda84af25932de7"
+		},
+		{
+			"ImportPath": "github.com/calmh/logger",
+			"Rev": "f50d32b313bec2933a3e1049f7416a29f3413d29"
+		},
+		{
+			"ImportPath": "github.com/calmh/osext",
+			"Rev": "9bf61584e5f1f172e8766ddc9022d9c401faaa5e"
+		},
+		{
+			"ImportPath": "github.com/calmh/xdr",
+			"Rev": "ec3d404f43731551258977b38dd72cf557d00398"
 		},
 		{
 			"ImportPath": "github.com/juju/ratelimit",
@@ -47,7 +51,11 @@
 		},
 		{
 			"ImportPath": "github.com/syndtr/goleveldb/leveldb",
-			"Rev": "e1f2d2bdccd7c62f4d4a29aaf081bf1fc4404f91"
+			"Rev": "d8d1d2a5cc2d34c950dffa2f554525415d59f737"
+		},
+		{
+			"ImportPath": "github.com/syndtr/gosnappy/snappy",
+			"Rev": "ce8acff4829e0c2458a67ead32390ac0a381c862"
 		},
 		{
 			"ImportPath": "github.com/vitrun/qart/coding",

Godeps/_workspace/src/code.google.com/p/go.crypto/blowfish/block.go

@@ -4,6 +4,22 @@
 
 package blowfish
 
+// getNextWord returns the next big-endian uint32 value from the byte slice
+// at the given position in a circular manner, updating the position.
+func getNextWord(b []byte, pos *int) uint32 {
+	var w uint32
+	j := *pos
+	for i := 0; i < 4; i++ {
+		w = w<<8 | uint32(b[j])
+		j++
+		if j >= len(b) {
+			j = 0
+		}
+	}
+	*pos = j
+	return w
+}
+
 // ExpandKey performs a key expansion on the given *Cipher. Specifically, it
 // performs the Blowfish algorithm's key schedule which sets up the *Cipher's
 // pi and substitution tables for calls to Encrypt. This is used, primarily,
@@ -12,6 +28,7 @@ package blowfish
 func ExpandKey(key []byte, c *Cipher) {
 	j := 0
 	for i := 0; i < 18; i++ {
+		// Using inlined getNextWord for performance.
 		var d uint32
 		for k := 0; k < 4; k++ {
 			d = d<<8 | uint32(key[j])
@@ -54,86 +71,44 @@ func ExpandKey(key []byte, c *Cipher) {
 func expandKeyWithSalt(key []byte, salt []byte, c *Cipher) {
 	j := 0
 	for i := 0; i < 18; i++ {
-		var d uint32
-		for k := 0; k < 4; k++ {
-			d = d<<8 | uint32(key[j])
-			j++
-			if j >= len(key) {
-				j = 0
-			}
-		}
-		c.p[i] ^= d
+		c.p[i] ^= getNextWord(key, &j)
 	}
 
 	j = 0
-	var expandedSalt [4]uint32
-	for i := range expandedSalt {
-		var d uint32
-		for k := 0; k < 4; k++ {
-			d = d<<8 | uint32(salt[j])
-			j++
-			if j >= len(salt) {
-				j = 0
-			}
-		}
-		expandedSalt[i] = d
-	}
-
 	var l, r uint32
 	for i := 0; i < 18; i += 2 {
-		l ^= expandedSalt[i&2]
-		r ^= expandedSalt[(i&2)+1]
+		l ^= getNextWord(salt, &j)
+		r ^= getNextWord(salt, &j)
 		l, r = encryptBlock(l, r, c)
 		c.p[i], c.p[i+1] = l, r
 	}
 
-	for i := 0; i < 256; i += 4 {
-		l ^= expandedSalt[2]
-		r ^= expandedSalt[3]
+	for i := 0; i < 256; i += 2 {
+		l ^= getNextWord(salt, &j)
+		r ^= getNextWord(salt, &j)
 		l, r = encryptBlock(l, r, c)
 		c.s0[i], c.s0[i+1] = l, r
-
-		l ^= expandedSalt[0]
-		r ^= expandedSalt[1]
-		l, r = encryptBlock(l, r, c)
-		c.s0[i+2], c.s0[i+3] = l, r
-
 	}
 
-	for i := 0; i < 256; i += 4 {
-		l ^= expandedSalt[2]
-		r ^= expandedSalt[3]
+	for i := 0; i < 256; i += 2 {
+		l ^= getNextWord(salt, &j)
+		r ^= getNextWord(salt, &j)
 		l, r = encryptBlock(l, r, c)
 		c.s1[i], c.s1[i+1] = l, r
-
-		l ^= expandedSalt[0]
-		r ^= expandedSalt[1]
-		l, r = encryptBlock(l, r, c)
-		c.s1[i+2], c.s1[i+3] = l, r
 	}
 
-	for i := 0; i < 256; i += 4 {
-		l ^= expandedSalt[2]
-		r ^= expandedSalt[3]
+	for i := 0; i < 256; i += 2 {
+		l ^= getNextWord(salt, &j)
+		r ^= getNextWord(salt, &j)
 		l, r = encryptBlock(l, r, c)
 		c.s2[i], c.s2[i+1] = l, r
-
-		l ^= expandedSalt[0]
-		r ^= expandedSalt[1]
-		l, r = encryptBlock(l, r, c)
-		c.s2[i+2], c.s2[i+3] = l, r
 	}
 
-	for i := 0; i < 256; i += 4 {
-		l ^= expandedSalt[2]
-		r ^= expandedSalt[3]
+	for i := 0; i < 256; i += 2 {
+		l ^= getNextWord(salt, &j)
+		r ^= getNextWord(salt, &j)
 		l, r = encryptBlock(l, r, c)
 		c.s3[i], c.s3[i+1] = l, r
-
-		l ^= expandedSalt[0]
-		r ^= expandedSalt[1]
-		l, r = encryptBlock(l, r, c)
-		c.s3[i+2], c.s3[i+3] = l, r
 	}
 }
 
@@ -182,9 +157,3 @@ func decryptBlock(l, r uint32, c *Cipher) (uint32, uint32) {
 	xr ^= c.p[0]
 	return xr, xl
 }
-
-func zero(x []uint32) {
-	for i := range x {
-		x[i] = 0
-	}
-}

Godeps/_workspace/src/code.google.com/p/go.crypto/blowfish/blowfish_test.go

@@ -4,9 +4,7 @@
 
 package blowfish
 
-import (
-	"testing"
-)
+import "testing"
 
 type CryptTest struct {
 	key []byte
@@ -202,3 +200,75 @@ func TestSaltedCipherKeyLength(t *testing.T) {
 		t.Errorf("NewSaltedCipher with long key, gave error %#v", err)
 	}
 }
+
+// Test vectors generated with Blowfish from OpenSSH.
+var saltedVectors = [][8]byte{
+	{0x0c, 0x82, 0x3b, 0x7b, 0x8d, 0x01, 0x4b, 0x7e},
+	{0xd1, 0xe1, 0x93, 0xf0, 0x70, 0xa6, 0xdb, 0x12},
+	{0xfc, 0x5e, 0xba, 0xde, 0xcb, 0xf8, 0x59, 0xad},
+	{0x8a, 0x0c, 0x76, 0xe7, 0xdd, 0x2c, 0xd3, 0xa8},
+	{0x2c, 0xcb, 0x7b, 0xee, 0xac, 0x7b, 0x7f, 0xf8},
+	{0xbb, 0xf6, 0x30, 0x6f, 0xe1, 0x5d, 0x62, 0xbf},
+	{0x97, 0x1e, 0xc1, 0x3d, 0x3d, 0xe0, 0x11, 0xe9},
+	{0x06, 0xd7, 0x4d, 0xb1, 0x80, 0xa3, 0xb1, 0x38},
+	{0x67, 0xa1, 0xa9, 0x75, 0x0e, 0x5b, 0xc6, 0xb4},
+	{0x51, 0x0f, 0x33, 0x0e, 0x4f, 0x67, 0xd2, 0x0c},
+	{0xf1, 0x73, 0x7e, 0xd8, 0x44, 0xea, 0xdb, 0xe5},
+	{0x14, 0x0e, 0x16, 0xce, 0x7f, 0x4a, 0x9c, 0x7b},
+	{0x4b, 0xfe, 0x43, 0xfd, 0xbf, 0x36, 0x04, 0x47},
+	{0xb1, 0xeb, 0x3e, 0x15, 0x36, 0xa7, 0xbb, 0xe2},
+	{0x6d, 0x0b, 0x41, 0xdd, 0x00, 0x98, 0x0b, 0x19},
+	{0xd3, 0xce, 0x45, 0xce, 0x1d, 0x56, 0xb7, 0xfc},
+	{0xd9, 0xf0, 0xfd, 0xda, 0xc0, 0x23, 0xb7, 0x93},
+	{0x4c, 0x6f, 0xa1, 0xe4, 0x0c, 0xa8, 0xca, 0x57},
+	{0xe6, 0x2f, 0x28, 0xa7, 0x0c, 0x94, 0x0d, 0x08},
+	{0x8f, 0xe3, 0xf0, 0xb6, 0x29, 0xe3, 0x44, 0x03},
+	{0xff, 0x98, 0xdd, 0x04, 0x45, 0xb4, 0x6d, 0x1f},
+	{0x9e, 0x45, 0x4d, 0x18, 0x40, 0x53, 0xdb, 0xef},
+	{0xb7, 0x3b, 0xef, 0x29, 0xbe, 0xa8, 0x13, 0x71},
+	{0x02, 0x54, 0x55, 0x41, 0x8e, 0x04, 0xfc, 0xad},
+	{0x6a, 0x0a, 0xee, 0x7c, 0x10, 0xd9, 0x19, 0xfe},
+	{0x0a, 0x22, 0xd9, 0x41, 0xcc, 0x23, 0x87, 0x13},
+	{0x6e, 0xff, 0x1f, 0xff, 0x36, 0x17, 0x9c, 0xbe},
+	{0x79, 0xad, 0xb7, 0x40, 0xf4, 0x9f, 0x51, 0xa6},
+	{0x97, 0x81, 0x99, 0xa4, 0xde, 0x9e, 0x9f, 0xb6},
+	{0x12, 0x19, 0x7a, 0x28, 0xd0, 0xdc, 0xcc, 0x92},
+	{0x81, 0xda, 0x60, 0x1e, 0x0e, 0xdd, 0x65, 0x56},
+	{0x7d, 0x76, 0x20, 0xb2, 0x73, 0xc9, 0x9e, 0xee},
+}
+
+func TestSaltedCipher(t *testing.T) {
+	var key, salt [32]byte
+	for i := range key {
+		key[i] = byte(i)
+		salt[i] = byte(i + 32)
+	}
+	for i, v := range saltedVectors {
+		c, err := NewSaltedCipher(key[:], salt[:i])
+		if err != nil {
+			t.Fatal(err)
+		}
+		var buf [8]byte
+		c.Encrypt(buf[:], buf[:])
+		if v != buf {
+			t.Errorf("%d: expected %x, got %x", i, v, buf)
+		}
+	}
+}
+
+func BenchmarkExpandKeyWithSalt(b *testing.B) {
+	key := make([]byte, 32)
+	salt := make([]byte, 16)
+	c, _ := NewCipher(key)
+	for i := 0; i < b.N; i++ {
+		expandKeyWithSalt(key, salt, c)
+	}
+}
+
+func BenchmarkExpandKey(b *testing.B) {
+	key := make([]byte, 32)
+	c, _ := NewCipher(key)
+	for i := 0; i < b.N; i++ {
+		ExpandKey(key, c)
+	}
+}

Godeps/_workspace/src/code.google.com/p/go.crypto/blowfish/cipher.go

@@ -40,8 +40,11 @@ func NewCipher(key []byte) (*Cipher, error) {
 // NewSaltedCipher creates a returns a Cipher that folds a salt into its key
 // schedule. For most purposes, NewCipher, instead of NewSaltedCipher, is
 // sufficient and desirable. For bcrypt compatiblity, the key can be over 56
-// bytes. Only the first 16 bytes of salt are used.
+// bytes.
 func NewSaltedCipher(key, salt []byte) (*Cipher, error) {
+	if len(salt) == 0 {
+		return NewCipher(key)
+	}
 	var result Cipher
 	if k := len(key); k < 1 {
 		return nil, KeySizeError(k)

Godeps/_workspace/src/code.google.com/p/go.text/transform/transform.go

@@ -9,6 +9,7 @@
 package transform
 
 import (
+	"bytes"
 	"errors"
 	"io"
 	"unicode/utf8"
@@ -127,7 +128,7 @@ func (r *Reader) Read(p []byte) (int, error) {
 				// cannot read more bytes into src.
 				r.transformComplete = r.err != nil
 				continue
-			case err == ErrShortDst && r.dst1 != 0:
+			case err == ErrShortDst && (r.dst1 != 0 || n != 0):
 				// Make room in dst by copying out, and try again.
 				continue
 			case err == ErrShortSrc && r.src1-r.src0 != len(r.src) && r.err == nil:
@@ -210,7 +211,7 @@ func (w *Writer) Write(data []byte) (n int, err error) {
 			n += nSrc
 		}
 		switch {
-		case err == ErrShortDst && nDst > 0:
+		case err == ErrShortDst && (nDst > 0 || nSrc > 0):
 		case err == ErrShortSrc && len(src) < len(w.src):
 			m := copy(w.src, src)
 			// If w.n > 0, bytes from data were already copied to w.src and n
@@ -467,30 +468,125 @@ func (t removeF) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err err
 	return
 }
 
-// Bytes returns a new byte slice with the result of converting b using t.
-// If any unrecoverable error occurs it returns nil.
-func Bytes(t Transformer, b []byte) []byte {
-	out := make([]byte, len(b))
-	n := 0
-	for {
-		nDst, nSrc, err := t.Transform(out[n:], b, true)
-		n += nDst
-		if err == nil {
-			return out[:n]
-		} else if err != ErrShortDst {
-			return nil
-		}
-		b = b[nSrc:]
+// grow returns a new []byte that is longer than b, and copies the first n bytes
+// of b to the start of the new slice.
+func grow(b []byte, n int) []byte {
+	m := len(b)
+	if m <= 256 {
+		m *= 2
+	} else {
+		m += m >> 1
+	}
+	buf := make([]byte, m)
+	copy(buf, b[:n])
+	return buf
+}
 
-		// Grow the destination buffer.
-		sz := len(out)
-		if sz <= 256 {
-			sz *= 2
-		} else {
-			sz += sz >> 1
+const initialBufSize = 128
+
+// String returns a string with the result of converting s[:n] using t, where
+// n <= len(s). If err == nil, n will be len(s).
+func String(t Transformer, s string) (result string, n int, err error) {
+	if s == "" {
+		return "", 0, nil
+	}
+
+	// Allocate only once. Note that both dst and src escape when passed to
+	// Transform.
+	buf := [2 * initialBufSize]byte{}
+	dst := buf[:initialBufSize:initialBufSize]
+	src := buf[initialBufSize : 2*initialBufSize]
+
+	// Avoid allocation if the transformed string is identical to the original.
+	// After this loop, pDst will point to the furthest point in s for which it
+	// could be detected that t gives equal results, src[:nSrc] will
+	// indicated the last processed chunk of s for which the output is not equal
+	// and dst[:nDst] will be the transform of this chunk.
+	var nDst, nSrc int
+	pDst := 0 // Used as index in both src and dst in this loop.
+	for {
+		n := copy(src, s[pDst:])
+		nDst, nSrc, err = t.Transform(dst, src[:n], pDst+n == len(s))
+
+		// Note 1: we will not enter the loop with pDst == len(s) and we will
+		// not end the loop with it either. So if nSrc is 0, this means there is
+		// some kind of error from which we cannot recover given the current
+		// buffer sizes. We will give up in this case.
+		// Note 2: it is not entirely correct to simply do a bytes.Equal as
+		// a Transformer may buffer internally. It will work in most cases,
+		// though, and no harm is done if it doesn't work.
+		// TODO:  let transformers implement an optional Spanner interface, akin
+		// to norm's QuickSpan. This would even allow us to avoid any allocation.
+		if nSrc == 0 || !bytes.Equal(dst[:nDst], src[:nSrc]) {
+			break
+		}
+
+		if pDst += nDst; pDst == len(s) {
+			return s, pDst, nil
+		}
+	}
+
+	// Move the bytes seen so far to dst.
+	pSrc := pDst + nSrc
+	if pDst+nDst <= initialBufSize {
+		copy(dst[pDst:], dst[:nDst])
+	} else {
+		b := make([]byte, len(s)+nDst-nSrc)
+		copy(b[pDst:], dst[:nDst])
+		dst = b
+	}
+	copy(dst, s[:pDst])
+	pDst += nDst
+
+	if err != nil && err != ErrShortDst && err != ErrShortSrc {
+		return string(dst[:pDst]), pSrc, err
+	}
+
+	// Complete the string with the remainder.
+	for {
+		n := copy(src, s[pSrc:])
+		nDst, nSrc, err = t.Transform(dst[pDst:], src[:n], pSrc+n == len(s))
+		pDst += nDst
+		pSrc += nSrc
+
+		switch err {
+		case nil:
+			if pSrc == len(s) {
+				return string(dst[:pDst]), pSrc, nil
+			}
+		case ErrShortDst:
+			// Do not grow as long as we can make progress. This may avoid
+			// excessive allocations.
+			if nDst == 0 {
+				dst = grow(dst, pDst)
+			}
+		case ErrShortSrc:
+			if nSrc == 0 {
+				src = grow(src, 0)
+			}
+		default:
+			return string(dst[:pDst]), pSrc, err
+		}
+	}
+}
+
+// Bytes returns a new byte slice with the result of converting b[:n] using t,
+// where n <= len(b). If err == nil, n will be len(b).
+func Bytes(t Transformer, b []byte) (result []byte, n int, err error) {
+	dst := make([]byte, len(b))
+	pDst, pSrc := 0, 0
+	for {
+		nDst, nSrc, err := t.Transform(dst[pDst:], b[pSrc:], true)
+		pDst += nDst
+		pSrc += nSrc
+		if err != ErrShortDst {
+			return dst[:pDst], pSrc, err
+		}
+
+		// Grow the destination buffer, but do not grow as long as we can make
+		// progress. This may avoid excessive allocations.
+		if nDst == 0 {
+			dst = grow(dst, pDst)
 		}
-		out2 := make([]byte, sz)
-		copy(out2, out[:n])
-		out = out2
 	}
 }

Godeps/_workspace/src/code.google.com/p/go.text/transform/transform_test.go

@@ -12,6 +12,7 @@ import (
 	"strconv"
 	"strings"
 	"testing"
+	"time"
 	"unicode/utf8"
 )
 
@@ -132,6 +133,43 @@ func (e rleEncode) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err e
 	return nDst, nSrc, nil
 }
 
+// trickler consumes all input bytes, but writes a single byte at a time to dst.
+type trickler []byte
+
+func (t *trickler) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err error) {
+	*t = append(*t, src...)
+	if len(*t) == 0 {
+		return 0, 0, nil
+	}
+	if len(dst) == 0 {
+		return 0, len(src), ErrShortDst
+	}
+	dst[0] = (*t)[0]
+	*t = (*t)[1:]
+	if len(*t) > 0 {
+		err = ErrShortDst
+	}
+	return 1, len(src), err
+}
+
+// delayedTrickler is like trickler, but delays writing output to dst. This is
+// highly unlikely to be relevant in practice, but it seems like a good idea
+// to have some tolerance as long as progress can be detected.
+type delayedTrickler []byte
+
+func (t *delayedTrickler) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err error) {
+	if len(*t) > 0 && len(dst) > 0 {
+		dst[0] = (*t)[0]
+		*t = (*t)[1:]
+		nDst = 1
+	}
+	*t = append(*t, src...)
+	if len(*t) > 0 {
+		err = ErrShortDst
+	}
+	return nDst, len(src), err
+}
+
 type testCase struct {
 	desc     string
 	t        Transformer
@@ -170,6 +208,15 @@ func (c chain) String() string {
 }
 
 var testCases = []testCase{
+	{
+		desc:    "empty",
+		t:       lowerCaseASCII{},
+		src:     "",
+		dstSize: 100,
+		srcSize: 100,
+		wantStr: "",
+	},
+
 	{
 		desc:    "basic",
 		t:       lowerCaseASCII{},
@@ -378,6 +425,24 @@ var testCases = []testCase{
 		ioSize:  10,
 		wantStr: "4a6b2b4c4d1d",
 	},
+
+	{
+		desc:    "trickler",
+		t:       &trickler{},
+		src:     "abcdefghijklm",
+		dstSize: 3,
+		srcSize: 15,
+		wantStr: "abcdefghijklm",
+	},
+
+	{
+		desc:    "delayedTrickler",
+		t:       &delayedTrickler{},
+		src:     "abcdefghijklm",
+		dstSize: 3,
+		srcSize: 15,
+		wantStr: "abcdefghijklm",
+	},
 }
 
 func TestReader(t *testing.T) {
@@ -685,7 +750,7 @@ func doTransform(tc testCase) (res string, iter int, err error) {
 		switch {
 		case err == nil && len(in) != 0:
 		case err == ErrShortSrc && nSrc > 0:
-		case err == ErrShortDst && nDst > 0:
+		case err == ErrShortDst && (nDst > 0 || nSrc > 0):
 		default:
 			return string(out), iter, err
 		}
@@ -875,27 +940,136 @@ func TestRemoveFunc(t *testing.T) {
 	}
 }
 
-func TestBytes(t *testing.T) {
+func testString(t *testing.T, f func(Transformer, string) (string, int, error)) {
 	for _, tt := range append(testCases, chainTests()...) {
 		if tt.desc == "allowStutter = true" {
 			// We don't have control over the buffer size, so we eliminate tests
 			// that depend on a specific buffer size being set.
 			continue
 		}
-		got := Bytes(tt.t, []byte(tt.src))
-		if tt.wantErr != nil {
-			if tt.wantErr != ErrShortDst && tt.wantErr != ErrShortSrc {
-				// Bytes should return nil for non-recoverable errors.
-				if g, w := (got == nil), (tt.wantErr != nil); g != w {
-					t.Errorf("%s:error: got %v; want %v", tt.desc, g, w)
-				}
-			}
-			// The output strings in the tests that expect an error will
-			// almost certainly not be the same as the result of Bytes.
+		reset(tt.t)
+		if tt.wantErr == ErrShortDst || tt.wantErr == ErrShortSrc {
+			// The result string will be different.
 			continue
 		}
-		if string(got) != tt.wantStr {
+		got, n, err := f(tt.t, tt.src)
+		if tt.wantErr != err {
+			t.Errorf("%s:error: got %v; want %v", tt.desc, err, tt.wantErr)
+		}
+		if got, want := err == nil, n == len(tt.src); got != want {
+			t.Errorf("%s:n: got %v; want %v", tt.desc, got, want)
+		}
+		if got != tt.wantStr {
 			t.Errorf("%s:string: got %q; want %q", tt.desc, got, tt.wantStr)
 		}
 	}
 }
+
+func TestBytes(t *testing.T) {
+	testString(t, func(z Transformer, s string) (string, int, error) {
+		b, n, err := Bytes(z, []byte(s))
+		return string(b), n, err
+	})
+}
+
+func TestString(t *testing.T) {
+	testString(t, String)
+
+	// Overrun the internal destination buffer.
+	for i, s := range []string{
+		strings.Repeat("a", initialBufSize-1),
+		strings.Repeat("a", initialBufSize+0),
+		strings.Repeat("a", initialBufSize+1),
+		strings.Repeat("A", initialBufSize-1),
+		strings.Repeat("A", initialBufSize+0),
+		strings.Repeat("A", initialBufSize+1),
+		strings.Repeat("A", 2*initialBufSize-1),
+		strings.Repeat("A", 2*initialBufSize+0),
+		strings.Repeat("A", 2*initialBufSize+1),
+		strings.Repeat("a", initialBufSize-2) + "A",
+		strings.Repeat("a", initialBufSize-1) + "A",
+		strings.Repeat("a", initialBufSize+0) + "A",
+		strings.Repeat("a", initialBufSize+1) + "A",
+	} {
+		got, _, _ := String(lowerCaseASCII{}, s)
+		if want := strings.ToLower(s); got != want {
+			t.Errorf("%d:dst buffer test: got %s (%d); want %s (%d)", i, got, len(got), want, len(want))
+		}
+	}
+
+	// Overrun the internal source buffer.
+	for i, s := range []string{
+		strings.Repeat("a", initialBufSize-1),
+		strings.Repeat("a", initialBufSize+0),
+		strings.Repeat("a", initialBufSize+1),
+		strings.Repeat("a", 2*initialBufSize+1),
+		strings.Repeat("a", 2*initialBufSize+0),
+		strings.Repeat("a", 2*initialBufSize+1),
+	} {
+		got, _, _ := String(rleEncode{}, s)
+		if want := fmt.Sprintf("%da", len(s)); got != want {
+			t.Errorf("%d:src buffer test: got %s (%d); want %s (%d)", i, got, len(got), want, len(want))
+		}
+	}
+
+	// Test allocations for non-changing strings.
+	// Note we still need to allocate a single buffer.
+	for i, s := range []string{
+		"",
+		"123",
+		"123456789",
+		strings.Repeat("a", initialBufSize),
+		strings.Repeat("a", 10*initialBufSize),
+	} {
+		if n := testing.AllocsPerRun(5, func() { String(&lowerCaseASCII{}, s) }); n > 1 {
+			t.Errorf("%d: #allocs was %f; want 1", i, n)
+		}
+	}
+}
+
+// TestBytesAllocation tests that buffer growth stays limited with the trickler
+// transformer, which behaves oddly but within spec. In case buffer growth is
+// not correctly handled, the test will either panic with a failed allocation or
+// thrash. To ensure the tests terminate under the last condition, we time out
+// after some sufficiently long period of time.
+func TestBytesAllocation(t *testing.T) {
+	done := make(chan bool)
+	go func() {
+		in := bytes.Repeat([]byte{'a'}, 1000)
+		tr := trickler(make([]byte, 1))
+		Bytes(&tr, in)
+		done <- true
+	}()
+	select {
+	case <-done:
+	case <-time.After(3 * time.Second):
+		t.Error("time out, likely due to excessive allocation")
+	}
+}
+
+// TestStringAllocation tests that buffer growth stays limited with the trickler
+// transformer, which behaves oddly but within spec. In case buffer growth is
+// not correctly handled, the test will either panic with a failed allocation or
+// thrash. To ensure the tests terminate under the last condition, we time out
+// after some sufficiently long period of time.
+func TestStringAllocation(t *testing.T) {
+	done := make(chan bool)
+	go func() {
+		in := strings.Repeat("a", 1000)
+		tr := trickler(make([]byte, 1))
+		String(&tr, in)
+		done <- true
+	}()
+	select {
+	case <-done:
+	case <-time.After(3 * time.Second):
+		t.Error("time out, likely due to excessive allocation")
+	}
+}
+
+func BenchmarkStringLower(b *testing.B) {
+	in := strings.Repeat("a", 4096)
+	for i := 0; i < b.N; i++ {
+		String(&lowerCaseASCII{}, in)
+	}
+}

Godeps/_workspace/src/code.google.com/p/go.text/unicode/norm/maketables.go

@@ -11,7 +11,6 @@
 package main
 
 import (
-	"bufio"
 	"bytes"
 	"flag"
 	"fmt"
@@ -24,6 +23,8 @@ import (
 	"strconv"
 	"strings"
 	"unicode"
+
+	"code.google.com/p/go.text/internal/ucd"
 )
 
 func main() {
@@ -63,31 +64,7 @@ var localFiles = flag.Bool("local",
 
 var logger = log.New(os.Stderr, "", log.Lshortfile)
 
-// UnicodeData.txt has form:
-//	0037;DIGIT SEVEN;Nd;0;EN;;7;7;7;N;;;;;
-//	007A;LATIN SMALL LETTER Z;Ll;0;L;;;;;N;;;005A;;005A
-// See http://unicode.org/reports/tr44/ for full explanation
-// The fields:
-const (
-	FCodePoint = iota
-	FName
-	FGeneralCategory
-	FCanonicalCombiningClass
-	FBidiClass
-	FDecompMapping
-	FDecimalValue
-	FDigitValue
-	FNumericValue
-	FBidiMirrored
-	FUnicode1Name
-	FISOComment
-	FSimpleUppercaseMapping
-	FSimpleLowercaseMapping
-	FSimpleTitlecaseMapping
-	NumField
-
-	MaxChar = 0x10FFFF // anything above this shouldn't exist
-)
+const MaxChar = 0x10FFFF // anything above this shouldn't exist
 
 // Quick Check properties of runes allow us to quickly
 // determine whether a rune may occur in a normal form.
@@ -232,7 +209,7 @@ func openReader(file string) (input io.ReadCloser) {
 	return
 }
 
-func parseDecomposition(s string, skipfirst bool) (a []rune, e error) {
+func parseDecomposition(s string, skipfirst bool) (a []rune, err error) {
 	decomp := strings.Split(s, " ")
 	if len(decomp) > 0 && skipfirst {
 		decomp = decomp[1:]
@@ -247,56 +224,31 @@ func parseDecomposition(s string, skipfirst bool) (a []rune, e error) {
 	return a, nil
 }
 
-func parseCharacter(line string) {
-	field := strings.Split(line, ";")
-	if len(field) != NumField {
-		logger.Fatalf("%5s: %d fields (expected %d)\n", line, len(field), NumField)
-	}
-	x, err := strconv.ParseUint(field[FCodePoint], 16, 64)
-	point := int(x)
-	if err != nil {
-		logger.Fatalf("%.5s...: %s", line, err)
-	}
-	if point == 0 {
-		return // not interesting and we use 0 as unset
-	}
-	if point > MaxChar {
-		logger.Fatalf("%5s: Rune %X > MaxChar (%X)", line, point, MaxChar)
-		return
-	}
-	state := SNormal
-	switch {
-	case strings.Index(field[FName], ", First>") > 0:
-		state = SFirst
-	case strings.Index(field[FName], ", Last>") > 0:
-		state = SLast
-	}
-	firstChar := lastChar + 1
-	lastChar = rune(point)
-	if state != SLast {
-		firstChar = lastChar
-	}
-	x, err = strconv.ParseUint(field[FCanonicalCombiningClass], 10, 64)
-	if err != nil {
-		logger.Fatalf("%U: bad ccc field: %s", int(x), err)
-	}
-	ccc := uint8(x)
-	decmap := field[FDecompMapping]
-	exp, e := parseDecomposition(decmap, false)
-	isCompat := false
-	if e != nil {
-		if len(decmap) > 0 {
-			exp, e = parseDecomposition(decmap, true)
-			if e != nil {
-				logger.Fatalf(`%U: bad decomp |%v|: "%s"`, int(x), decmap, e)
+func loadUnicodeData() {
+	f := openReader("UnicodeData.txt")
+	defer f.Close()
+	p := ucd.New(f)
+	for p.Next() {
+		r := p.Rune(ucd.CodePoint)
+		char := &chars[r]
+
+		char.ccc = uint8(p.Uint(ucd.CanonicalCombiningClass))
+		decmap := p.String(ucd.DecompMapping)
+
+		exp, err := parseDecomposition(decmap, false)
+		isCompat := false
+		if err != nil {
+			if len(decmap) > 0 {
+				exp, err = parseDecomposition(decmap, true)
+				if err != nil {
+					logger.Fatalf(`%U: bad decomp |%v|: "%s"`, r, decmap, err)
+				}
+				isCompat = true
 			}
-			isCompat = true
 		}
-	}
-	for i := firstChar; i <= lastChar; i++ {
-		char := &chars[i]
-		char.name = field[FName]
-		char.codePoint = i
+
+		char.name = p.String(ucd.Name)
+		char.codePoint = r
 		char.forms[FCompatibility].decomp = exp
 		if !isCompat {
 			char.forms[FCanonical].decomp = exp
@@ -306,24 +258,9 @@ func parseCharacter(line string) {
 		if len(decmap) > 0 {
 			char.forms[FCompatibility].decomp = exp
 		}
-		char.ccc = ccc
-		char.state = SMissing
-		if i == lastChar {
-			char.state = state
-		}
 	}
-	return
-}
-
-func loadUnicodeData() {
-	f := openReader("UnicodeData.txt")
-	defer f.Close()
-	scanner := bufio.NewScanner(f)
-	for scanner.Scan() {
-		parseCharacter(scanner.Text())
-	}
-	if scanner.Err() != nil {
-		logger.Fatal(scanner.Err())
+	if err := p.Err(); err != nil {
+		logger.Fatal(err)
 	}
 }
 
@@ -354,47 +291,22 @@ func compactCCC() {
 	}
 }
 
-var singlePointRe = regexp.MustCompile(`^([0-9A-F]+) *$`)
-
 // CompositionExclusions.txt has form:
 // 0958    # ...
 // See http://unicode.org/reports/tr44/ for full explanation
-func parseExclusion(line string) int {
-	comment := strings.Index(line, "#")
-	if comment >= 0 {
-		line = line[0:comment]
-	}
-	if len(line) == 0 {
-		return 0
-	}
-	matches := singlePointRe.FindStringSubmatch(line)
-	if len(matches) != 2 {
-		logger.Fatalf("%s: %d matches (expected 1)\n", line, len(matches))
-	}
-	point, err := strconv.ParseUint(matches[1], 16, 64)
-	if err != nil {
-		logger.Fatalf("%.5s...: %s", line, err)
-	}
-	return int(point)
-}
-
 func loadCompositionExclusions() {
 	f := openReader("CompositionExclusions.txt")
 	defer f.Close()
-	scanner := bufio.NewScanner(f)
-	for scanner.Scan() {
-		point := parseExclusion(scanner.Text())
-		if point == 0 {
-			continue
-		}
-		c := &chars[point]
+	p := ucd.New(f)
+	for p.Next() {
+		c := &chars[p.Rune(0)]
 		if c.excludeInComp {
 			logger.Fatalf("%U: Duplicate entry in exclusions.", c.codePoint)
 		}
 		c.excludeInComp = true
 	}
-	if scanner.Err() != nil {
-		log.Fatal(scanner.Err())
+	if e := p.Err(); e != nil {
+		logger.Fatal(e)
 	}
 }
 
@@ -988,8 +900,6 @@ func verifyComputed() {
 	}
 }
 
-var qcRe = regexp.MustCompile(`([0-9A-F\.]+) *; (NF.*_QC); ([YNM]) #.*`)
-
 // Use values in DerivedNormalizationProps.txt to compare against the
 // values we computed.
 // DerivedNormalizationProps.txt has form:
@@ -999,27 +909,13 @@ var qcRe = regexp.MustCompile(`([0-9A-F\.]+) *; (NF.*_QC); ([YNM]) #.*`)
 func testDerived() {
 	f := openReader("DerivedNormalizationProps.txt")
 	defer f.Close()
-	scanner := bufio.NewScanner(f)
-	for scanner.Scan() {
-		line := scanner.Text()
-		qc := qcRe.FindStringSubmatch(line)
-		if qc == nil {
-			continue
-		}
-		rng := strings.Split(qc[1], "..")
-		i, err := strconv.ParseUint(rng[0], 16, 64)
-		if err != nil {
-			log.Fatal(err)
-		}
-		j := i
-		if len(rng) > 1 {
-			j, err = strconv.ParseUint(rng[1], 16, 64)
-			if err != nil {
-				log.Fatal(err)
-			}
-		}
+	p := ucd.New(f)
+	for p.Next() {
+		r := p.Rune(0)
+		c := &chars[r]
+
 		var ftype, mode int
-		qt := strings.TrimSpace(qc[2])
+		qt := p.String(1)
 		switch qt {
 		case "NFC_QC":
 			ftype, mode = FCanonical, MComposed
@@ -1030,10 +926,10 @@ func testDerived() {
 		case "NFKD_QC":
 			ftype, mode = FCompatibility, MDecomposed
 		default:
-			log.Fatalf(`Unexpected quick check type "%s"`, qt)
+			continue
 		}
 		var qr QCResult
-		switch qc[3] {
+		switch p.String(2) {
 		case "Y":
 			qr = QCYes
 		case "N":
@@ -1041,27 +937,15 @@ func testDerived() {
 		case "M":
 			qr = QCMaybe
 		default:
-			log.Fatalf(`Unexpected quick check value "%s"`, qc[3])
+			log.Fatalf(`Unexpected quick check value "%s"`, p.String(2))
 		}
-		var lastFailed bool
-		// Verify current
-		for ; i <= j; i++ {
-			c := &chars[int(i)]
-			c.forms[ftype].verified[mode] = true
-			curqr := c.forms[ftype].quickCheck[mode]
-			if curqr != qr {
-				if !lastFailed {
-					logger.Printf("%s: %.4X..%.4X -- %s\n",
-						qt, int(i), int(j), line[0:50])
-				}
-				logger.Printf("%U: FAILED %s (was %v need %v)\n",
-					int(i), qt, curqr, qr)
-				lastFailed = true
-			}
+		if got := c.forms[ftype].quickCheck[mode]; got != qr {
+			logger.Printf("%U: FAILED %s (was %v need %v)\n", r, qt, got, qr)
 		}
+		c.forms[ftype].verified[mode] = true
 	}
-	if scanner.Err() != nil {
-		logger.Fatal(scanner.Err())
+	if err := p.Err(); err != nil {
+		logger.Fatal(err)
 	}
 	// Any unspecified value must be QCYes. Verify this.
 	for i, c := range chars {

Godeps/_workspace/src/github.com/AudriusButkevicius/lfu-go/LICENSE

@@ -0,0 +1,19 @@
+Copyright (C) 2012 Dave Grijalva
+
+Permission is hereby granted, free of charge, to any person obtaining a 
+copy of this software and associated documentation files (the "Software"), 
+to deal in the Software without restriction, including without limitation 
+the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+and/or sell copies of the Software, and to permit persons to whom the 
+Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included 
+in all copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS 
+OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 
+THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 
+THE SOFTWARE.

Godeps/_workspace/src/github.com/AudriusButkevicius/lfu-go/README.md

@@ -0,0 +1,19 @@
+A simple LFU cache for golang.  Based on the paper [An O(1) algorithm for implementing the LFU cache eviction scheme](http://dhruvbird.com/lfu.pdf).
+
+Usage:
+
+```go
+import "github.com/dgrijalva/lfu-go"
+
+// Make a new thing
+c := lfu.New()
+
+// Set some values
+c.Set("myKey", myValue)
+
+// Retrieve some values
+myValue = c.Get("myKey")
+
+// Evict some values
+c.Evict(1)
+```

Godeps/_workspace/src/github.com/AudriusButkevicius/lfu-go/lfu.go

@@ -0,0 +1,156 @@
+package lfu
+
+import (
+	"container/list"
+	"sync"
+)
+
+type Eviction struct {
+	Key   string
+	Value interface{}
+}
+
+type Cache struct {
+	// If len > UpperBound, cache will automatically evict
+	// down to LowerBound.  If either value is 0, this behavior
+	// is disabled.
+	UpperBound      int
+	LowerBound      int
+	values          map[string]*cacheEntry
+	freqs           *list.List
+	len             int
+	lock            *sync.Mutex
+	EvictionChannel chan<- Eviction
+}
+
+type cacheEntry struct {
+	key      string
+	value    interface{}
+	freqNode *list.Element
+}
+
+type listEntry struct {
+	entries map[*cacheEntry]byte
+	freq    int
+}
+
+func New() *Cache {
+	c := new(Cache)
+	c.values = make(map[string]*cacheEntry)
+	c.freqs = list.New()
+	c.lock = new(sync.Mutex)
+	return c
+}
+
+func (c *Cache) Get(key string) interface{} {
+	c.lock.Lock()
+	defer c.lock.Unlock()
+	if e, ok := c.values[key]; ok {
+		c.increment(e)
+		return e.value
+	}
+	return nil
+}
+
+func (c *Cache) Set(key string, value interface{}) {
+	c.lock.Lock()
+	defer c.lock.Unlock()
+	if e, ok := c.values[key]; ok {
+		// value already exists for key.  overwrite
+		e.value = value
+		c.increment(e)
+	} else {
+		// value doesn't exist.  insert
+		e := new(cacheEntry)
+		e.key = key
+		e.value = value
+		c.values[key] = e
+		c.increment(e)
+		c.len++
+		// bounds mgmt
+		if c.UpperBound > 0 && c.LowerBound > 0 {
+			if c.len > c.UpperBound {
+				c.evict(c.len - c.LowerBound)
+			}
+		}
+	}
+}
+
+func (c *Cache) Len() int {
+	c.lock.Lock()
+	defer c.lock.Unlock()
+	return c.len
+}
+
+func (c *Cache) Evict(count int) int {
+	c.lock.Lock()
+	defer c.lock.Unlock()
+	return c.evict(count)
+}
+
+func (c *Cache) evict(count int) int {
+	// No lock here so it can be called
+	// from within the lock (during Set)
+	var evicted int
+	for i := 0; i < count; {
+		if place := c.freqs.Front(); place != nil {
+			for entry, _ := range place.Value.(*listEntry).entries {
+				if i < count {
+					if c.EvictionChannel != nil {
+						c.EvictionChannel <- Eviction{
+							Key:   entry.key,
+							Value: entry.value,
+						}
+					}
+					delete(c.values, entry.key)
+					c.remEntry(place, entry)
+					evicted++
+					c.len--
+					i++
+				}
+			}
+		}
+	}
+	return evicted
+}
+
+func (c *Cache) increment(e *cacheEntry) {
+	currentPlace := e.freqNode
+	var nextFreq int
+	var nextPlace *list.Element
+	if currentPlace == nil {
+		// new entry
+		nextFreq = 1
+		nextPlace = c.freqs.Front()
+	} else {
+		// move up
+		nextFreq = currentPlace.Value.(*listEntry).freq + 1
+		nextPlace = currentPlace.Next()
+	}
+
+	if nextPlace == nil || nextPlace.Value.(*listEntry).freq != nextFreq {
+		// create a new list entry
+		li := new(listEntry)
+		li.freq = nextFreq
+		li.entries = make(map[*cacheEntry]byte)
+		if currentPlace != nil {
+			nextPlace = c.freqs.InsertAfter(li, currentPlace)
+		} else {
+			nextPlace = c.freqs.PushFront(li)
+		}
+	}
+	e.freqNode = nextPlace
+	nextPlace.Value.(*listEntry).entries[e] = 1
+	if currentPlace != nil {
+		// remove from current position
+		c.remEntry(currentPlace, e)
+	}
+}
+
+func (c *Cache) remEntry(place *list.Element, entry *cacheEntry) {
+	entries := place.Value.(*listEntry).entries
+	delete(entries, entry)
+	if len(entries) == 0 {
+		c.freqs.Remove(place)
+	}
+}

Godeps/_workspace/src/github.com/AudriusButkevicius/lfu-go/lfu_test.go

@@ -0,0 +1,68 @@
+package lfu
+
+import (
+	"fmt"
+	"testing"
+)
+
+func TestLFU(t *testing.T) {
+	c := New()
+	c.Set("a", "a")
+	if v := c.Get("a"); v != "a" {
+		t.Errorf("Value was not saved: %v != 'a'", v)
+	}
+	if l := c.Len(); l != 1 {
+		t.Errorf("Length was not updated: %v != 1", l)
+	}
+
+	c.Set("b", "b")
+	if v := c.Get("b"); v != "b" {
+		t.Errorf("Value was not saved: %v != 'b'", v)
+	}
+	if l := c.Len(); l != 2 {
+		t.Errorf("Length was not updated: %v != 2", l)
+	}
+
+	c.Get("a")
+	evicted := c.Evict(1)
+	if v := c.Get("a"); v != "a" {
+		t.Errorf("Value was improperly evicted: %v != 'a'", v)
+	}
+	if v := c.Get("b"); v != nil {
+		t.Errorf("Value was not evicted: %v", v)
+	}
+	if l := c.Len(); l != 1 {
+		t.Errorf("Length was not updated: %v != 1", l)
+	}
+	if evicted != 1 {
+		t.Errorf("Number of evicted items is wrong: %v != 1", evicted)
+	}
+}
+
+func TestBoundsMgmt(t *testing.T) {
+	c := New()
+	c.UpperBound = 10
+	c.LowerBound = 5
+
+	for i := 0; i < 100; i++ {
+		c.Set(fmt.Sprintf("%v", i), i)
+	}
+	if c.Len() > 10 {
+		t.Errorf("Bounds management failed to evict properly: %v", c.Len())
+	}
+}
+
+func TestEviction(t *testing.T) {
+	ch := make(chan Eviction, 1)
+
+	c := New()
+	c.EvictionChannel = ch
+	c.Set("a", "b")
+	c.Evict(1)
+
+	ev := <-ch
+
+	if ev.Key != "a" || ev.Value.(string) != "b" {
+		t.Error("Incorrect item")
+	}
+}

Godeps/_workspace/src/github.com/bkaradzic/go-lz4/.gitignore

@@ -0,0 +1 @@
+/lz4-example/lz4-example

Godeps/_workspace/src/github.com/bkaradzic/go-lz4/.travis.yml

@@ -0,0 +1,7 @@
+language: go
+
+go:
+  - 1.1
+  - 1.2
+  - 1.3
+  - tip

Godeps/_workspace/src/github.com/bkaradzic/go-lz4/LICENSE

@@ -0,0 +1,24 @@
+Copyright 2011-2012 Branimir Karadzic. All rights reserved.
+Copyright 2013 Damian Gryski. All rights reserved.
+
+Redistribution and use in source and binary forms, with or without modification,
+are permitted provided that the following conditions are met:
+
+   1. Redistributions of source code must retain the above copyright notice, this
+      list of conditions and the following disclaimer.
+
+   2. Redistributions in binary form must reproduce the above copyright notice,
+      this list of conditions and the following disclaimer in the documentation
+      and/or other materials provided with the distribution.
+
+THIS SOFTWARE IS PROVIDED BY COPYRIGHT HOLDER ``AS IS'' AND ANY EXPRESS OR
+IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT
+SHALL COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
+WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE
+OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
+THE POSSIBILITY OF SUCH DAMAGE.
+

Godeps/_workspace/src/github.com/bkaradzic/go-lz4/README.md

@@ -0,0 +1,71 @@
+go-lz4
+======
+
+go-lz4 is port of LZ4 lossless compression algorithm to Go. The original C code
+is located at:
+
+https://code.google.com/p/lz4/
+
+Status
+------
+[![Build Status](https://secure.travis-ci.org/bkaradzic/go-lz4.png)](http://travis-ci.org/bkaradzic/go-lz4)  
+[![GoDoc](https://godoc.org/github.com/bkaradzic/go-lz4?status.png)](https://godoc.org/github.com/bkaradzic/go-lz4)
+
+Usage
+-----
+
+    go get github.com/bkaradzic/go-lz4
+
+    import "github.com/bkaradzic/go-lz4"
+
+The package name is `lz4`
+
+Notes
+-----
+
+* go-lz4 saves a uint32 with the original uncompressed length at the beginning
+  of the encoded buffer.  They may get in the way of interoperability with
+  other implementations.
+
+Contributors
+------------
+
+Damian Gryski ([@dgryski](https://github.com/dgryski))  
+Dustin Sallings ([@dustin](https://github.com/dustin))
+
+Contact
+-------
+
+[@bkaradzic](https://twitter.com/bkaradzic)  
+http://www.stuckingeometry.com
+
+Project page  
+https://github.com/bkaradzic/go-lz4
+
+License
+-------
+
+Copyright 2011-2012 Branimir Karadzic. All rights reserved.  
+Copyright 2013 Damian Gryski. All rights reserved.
+
+Redistribution and use in source and binary forms, with or without modification,
+are permitted provided that the following conditions are met:
+
+   1. Redistributions of source code must retain the above copyright notice, this
+      list of conditions and the following disclaimer.
+
+   2. Redistributions in binary form must reproduce the above copyright notice,
+      this list of conditions and the following disclaimer in the documentation
+      and/or other materials provided with the distribution.
+
+THIS SOFTWARE IS PROVIDED BY COPYRIGHT HOLDER ``AS IS'' AND ANY EXPRESS OR
+IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT
+SHALL COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
+WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE
+OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
+THE POSSIBILITY OF SUCH DAMAGE.
+

Godeps/_workspace/src/github.com/bkaradzic/go-lz4/fuzzer/main.go

@@ -0,0 +1,74 @@
+package main
+
+import (
+	"math/rand"
+
+	"github.com/bkaradzic/go-lz4"
+
+	// lz4's API matches snappy's, so we can easily see how it performs
+	// lz4 "code.google.com/p/snappy-go/snappy"
+)
+
+var input = `
+ADVENTURE I. A SCANDAL IN BOHEMIA
+
+I.
+
+To Sherlock Holmes she is always THE woman. I have seldom heard
+him mention her under any other name. In his eyes she eclipses
+and predominates the whole of her sex. It was not that he felt
+any emotion akin to love for Irene Adler. All emotions, and that
+one particularly, were abhorrent to his cold, precise but
+admirably balanced mind. He was, I take it, the most perfect
+reasoning and observing machine that the world has seen, but as a
+lover he would have placed himself in a false position. He never
+spoke of the softer passions, save with a gibe and a sneer. They
+were admirable things for the observer--excellent for drawing the
+veil from men's motives and actions. But for the trained reasoner
+to admit such intrusions into his own delicate and finely
+adjusted temperament was to introduce a distracting factor which
+might throw a doubt upon all his mental results. Grit in a
+sensitive instrument, or a crack in one of his own high-power
+lenses, would not be more disturbing than a strong emotion in a
+nature such as his. And yet there was but one woman to him, and
+that woman was the late Irene Adler, of dubious and questionable
+memory.
+
+I had seen little of Holmes lately. My marriage had drifted us
+away from each other. My own complete happiness, and the
+home-centred interests which rise up around the man who first
+finds himself master of his own establishment, were sufficient to
+absorb all my attention, while Holmes, who loathed every form of
+society with his whole Bohemian soul, remained in our lodgings in
+Baker Street, buried among his old books, and alternating from
+week to week between cocaine and ambition, the drowsiness of the
+drug, and the fierce energy of his own keen nature. He was still,
+as ever, deeply attracted by the study of crime, and occupied his
+immense faculties and extraordinary powers of observation in
+following out those clues, and clearing up those mysteries which
+had been abandoned as hopeless by the official police. From time
+to time I heard some vague account of his doings: of his summons
+to Odessa in the case of the Trepoff murder, of his clearing up
+of the singular tragedy of the Atkinson brothers at Trincomalee,
+and finally of the mission which he had accomplished so
+delicately and successfully for the reigning family of Holland.
+Beyond these signs of his activity, however, which I merely
+shared with all the readers of the daily press, I knew little of
+my former friend and companion.
+`
+
+func main() {
+
+	compressed, _ := lz4.Encode(nil, []byte(input))
+
+	modified := make([]byte, len(compressed))
+
+	for {
+		copy(modified, compressed)
+		for i := 0; i < 100; i++ {
+			modified[rand.Intn(len(compressed)-4)+4] = byte(rand.Intn(256))
+		}
+		lz4.Decode(nil, modified)
+	}
+
+}

Godeps/_workspace/src/github.com/bkaradzic/go-lz4/lz4-example/main.go

@@ -0,0 +1,94 @@
+/*
+ * Copyright 2011 Branimir Karadzic. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without modification,
+ * are permitted provided that the following conditions are met:
+ *
+ *    1. Redistributions of source code must retain the above copyright notice, this
+ *       list of conditions and the following disclaimer.
+ *
+ *    2. Redistributions in binary form must reproduce the above copyright notice,
+ *       this list of conditions and the following disclaimer in the documentation
+ *       and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY COPYRIGHT HOLDER ``AS IS'' AND ANY EXPRESS OR
+ * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT
+ * SHALL COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
+ * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE
+ * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
+ * THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+package main
+
+import (
+	"flag"
+	"fmt"
+	"io/ioutil"
+	"log"
+	"os"
+	"runtime/pprof"
+
+	lz4 "github.com/bkaradzic/go-lz4"
+)
+
+var (
+	decompress = flag.Bool("d", false, "decompress")
+)
+
+func main() {
+
+	var optCPUProfile = flag.String("cpuprofile", "", "profile")
+	flag.Parse()
+
+	if *optCPUProfile != "" {
+		f, err := os.Create(*optCPUProfile)
+		if err != nil {
+			log.Fatal(err)
+		}
+		pprof.StartCPUProfile(f)
+		defer pprof.StopCPUProfile()
+	}
+
+	args := flag.Args()
+
+	var data []byte
+
+	if len(args) < 2 {
+		fmt.Print("Usage: lz4 [-d] <input> <output>\n")
+		os.Exit(1)
+	}
+
+	input, err := os.OpenFile(args[0], os.O_RDONLY, 0644)
+	if err != nil {
+		fmt.Printf("Failed to open input file %s\n", args[0])
+		os.Exit(1)
+	}
+	defer input.Close()
+
+	if *decompress {
+		data, _ = ioutil.ReadAll(input)
+		data, err = lz4.Decode(nil, data)
+		if err != nil {
+			fmt.Println("Failed to decode:", err)
+			return
+		}
+	} else {
+		data, _ = ioutil.ReadAll(input)
+		data, err = lz4.Encode(nil, data)
+		if err != nil {
+			fmt.Println("Failed to encode:", err)
+			return
+		}
+	}
+
+	err = ioutil.WriteFile(args[1], data, 0644)
+	if err != nil {
+		fmt.Printf("Failed to open output file %s\n", args[1])
+		os.Exit(1)
+	}
+}

Godeps/_workspace/src/github.com/bkaradzic/go-lz4/lz4_test.go

@@ -0,0 +1,63 @@
+package lz4
+
+import (
+	"bytes"
+	"io/ioutil"
+	"testing"
+)
+
+var testfile, _ = ioutil.ReadFile("testdata/pg1661.txt")
+
+func roundtrip(t *testing.T, input []byte) {
+
+	dst, err := Encode(nil, input)
+	if err != nil {
+		t.Errorf("got error during compression: %s", err)
+	}
+
+	output, err := Decode(nil, dst)
+
+	if err != nil {
+		t.Errorf("got error during decompress: %s", err)
+	}
+
+	if !bytes.Equal(output, input) {
+		t.Errorf("roundtrip failed")
+	}
+}
+
+func TestEmpty(t *testing.T) {
+	roundtrip(t, nil)
+}
+
+func TestLengths(t *testing.T) {
+
+	for i := 0; i < 1024; i++ {
+		roundtrip(t, testfile[:i])
+	}
+
+	for i := 1024; i < 4096; i += 23 {
+		roundtrip(t, testfile[:i])
+	}
+}
+
+func TestWords(t *testing.T) {
+	roundtrip(t, testfile)
+}
+
+func BenchmarkLZ4Encode(b *testing.B) {
+	for i := 0; i < b.N; i++ {
+		Encode(nil, testfile)
+	}
+}
+
+func BenchmarkLZ4Decode(b *testing.B) {
+
+	var compressed, _ = Encode(nil, testfile)
+
+	b.ResetTimer()
+
+	for i := 0; i < b.N; i++ {
+		Decode(nil, compressed)
+	}
+}

Godeps/_workspace/src/github.com/bkaradzic/go-lz4/reader.go

@@ -0,0 +1,194 @@
+/*
+ * Copyright 2011-2012 Branimir Karadzic. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without modification,
+ * are permitted provided that the following conditions are met:
+ *
+ *    1. Redistributions of source code must retain the above copyright notice, this
+ *       list of conditions and the following disclaimer.
+ *
+ *    2. Redistributions in binary form must reproduce the above copyright notice,
+ *       this list of conditions and the following disclaimer in the documentation
+ *       and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY COPYRIGHT HOLDER ``AS IS'' AND ANY EXPRESS OR
+ * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT
+ * SHALL COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
+ * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE
+ * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
+ * THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+package lz4
+
+import (
+	"encoding/binary"
+	"errors"
+	"io"
+)
+
+var (
+	// ErrCorrupt indicates the input was corrupt
+	ErrCorrupt = errors.New("corrupt input")
+)
+
+const (
+	mlBits  = 4
+	mlMask  = (1 << mlBits) - 1
+	runBits = 8 - mlBits
+	runMask = (1 << runBits) - 1
+)
+
+type decoder struct {
+	src  []byte
+	dst  []byte
+	spos uint32
+	dpos uint32
+	ref  uint32
+}
+
+func (d *decoder) readByte() (uint8, error) {
+	if int(d.spos) == len(d.src) {
+		return 0, io.EOF
+	}
+	b := d.src[d.spos]
+	d.spos++
+	return b, nil
+}
+
+func (d *decoder) getLen() (uint32, error) {
+
+	length := uint32(0)
+	ln, err := d.readByte()
+	if err != nil {
+		return 0, ErrCorrupt
+	}
+	for ln == 255 {
+		length += 255
+		ln, err = d.readByte()
+		if err != nil {
+			return 0, ErrCorrupt
+		}
+	}
+	length += uint32(ln)
+
+	return length, nil
+}
+
+func (d *decoder) cp(length, decr uint32) {
+
+	if int(d.ref+length) < int(d.dpos) {
+		copy(d.dst[d.dpos:], d.dst[d.ref:d.ref+length])
+	} else {
+		for ii := uint32(0); ii < length; ii++ {
+			d.dst[d.dpos+ii] = d.dst[d.ref+ii]
+		}
+	}
+	d.dpos += length
+	d.ref += length - decr
+}
+
+func (d *decoder) finish(err error) error {
+	if err == io.EOF {
+		return nil
+	}
+
+	return err
+}
+
+// Decode returns the decoded form of src.  The returned slice may be a
+// subslice of dst if it was large enough to hold the entire decoded block.
+func Decode(dst, src []byte) ([]byte, error) {
+
+	if len(src) < 4 {
+		return nil, ErrCorrupt
+	}
+
+	uncompressedLen := binary.LittleEndian.Uint32(src)
+
+	if uncompressedLen == 0 {
+		return nil, nil
+	}
+
+	if uncompressedLen > MaxInputSize {
+		return nil, ErrTooLarge
+	}
+
+	if dst == nil || len(dst) < int(uncompressedLen) {
+		dst = make([]byte, uncompressedLen)
+	}
+
+	d := decoder{src: src, dst: dst[:uncompressedLen], spos: 4}
+
+	decr := []uint32{0, 3, 2, 3}
+
+	for {
+		code, err := d.readByte()
+		if err != nil {
+			return d.dst, d.finish(err)
+		}
+
+		length := uint32(code >> mlBits)
+		if length == runMask {
+			ln, err := d.getLen()
+			if err != nil {
+				return nil, ErrCorrupt
+			}
+			length += ln
+		}
+
+		if int(d.spos+length) > len(d.src) {
+			return nil, ErrCorrupt
+		}
+
+		for ii := uint32(0); ii < length; ii++ {
+			d.dst[d.dpos+ii] = d.src[d.spos+ii]
+		}
+
+		d.spos += length
+		d.dpos += length
+
+		if int(d.spos) == len(d.src) {
+			return d.dst, nil
+		}
+
+		if int(d.spos+2) >= len(d.src) {
+			return nil, ErrCorrupt
+		}
+
+		back := uint32(d.src[d.spos]) | uint32(d.src[d.spos+1])<<8
+
+		if back > d.dpos {
+			return nil, ErrCorrupt
+		}
+
+		d.spos += 2
+		d.ref = d.dpos - back
+
+		length = uint32(code & mlMask)
+		if length == mlMask {
+			ln, err := d.getLen()
+			if err != nil {
+				return nil, ErrCorrupt
+			}
+			length += ln
+		}
+
+		literal := d.dpos - d.ref
+		if literal < 4 {
+			d.cp(4, decr[literal])
+		} else {
+			length += 4
+		}
+
+		if d.dpos+length > uncompressedLen {
+			return nil, ErrCorrupt
+		}
+
+		d.cp(length, 0)
+	}
+}

Godeps/_workspace/src/github.com/bkaradzic/go-lz4/writer.go

@@ -0,0 +1,188 @@
+/*
+ * Copyright 2011-2012 Branimir Karadzic. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without modification,
+ * are permitted provided that the following conditions are met:
+ *
+ *    1. Redistributions of source code must retain the above copyright notice, this
+ *       list of conditions and the following disclaimer.
+ *
+ *    2. Redistributions in binary form must reproduce the above copyright notice,
+ *       this list of conditions and the following disclaimer in the documentation
+ *       and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY COPYRIGHT HOLDER ``AS IS'' AND ANY EXPRESS OR
+ * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT
+ * SHALL COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
+ * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE
+ * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
+ * THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+package lz4
+
+import "encoding/binary"
+import "errors"
+
+const (
+	minMatch              = 4
+	hashLog               = 17
+	hashTableSize         = 1 << hashLog
+	hashShift             = (minMatch * 8) - hashLog
+	incompressible uint32 = 128
+	uninitHash            = 0x88888888
+
+	// MaxInputSize is the largest buffer than can be compressed in a single block
+	MaxInputSize = 0x7E000000
+)
+
+var (
+	// ErrTooLarge indicates the input buffer was too large
+	ErrTooLarge = errors.New("input too large")
+)
+
+type encoder struct {
+	src       []byte
+	dst       []byte
+	hashTable []uint32
+	pos       uint32
+	anchor    uint32
+	dpos      uint32
+}
+
+// CompressBound returns the maximum length of a lz4 block, given it's uncompressed length
+func CompressBound(isize int) int {
+	if isize > MaxInputSize {
+		return 0
+	}
+	return isize + ((isize) / 255) + 16 + 4
+}
+
+func (e *encoder) writeLiterals(length, mlLen, pos uint32) {
+
+	ln := length
+
+	var code byte
+	if ln > runMask-1 {
+		code = runMask
+	} else {
+		code = byte(ln)
+	}
+
+	if mlLen > mlMask-1 {
+		e.dst[e.dpos] = (code << mlBits) + byte(mlMask)
+	} else {
+		e.dst[e.dpos] = (code << mlBits) + byte(mlLen)
+	}
+	e.dpos++
+
+	if code == runMask {
+		ln -= runMask
+		for ; ln > 254; ln -= 255 {
+			e.dst[e.dpos] = 255
+			e.dpos++
+		}
+
+		e.dst[e.dpos] = byte(ln)
+		e.dpos++
+	}
+
+	for ii := uint32(0); ii < length; ii++ {
+		e.dst[e.dpos+ii] = e.src[pos+ii]
+	}
+
+	e.dpos += length
+}
+
+// Encode returns the encoded form of src.  The returned array may be a
+// sub-slice of dst if it was large enough to hold the entire output.
+func Encode(dst, src []byte) ([]byte, error) {
+
+	if len(src) >= MaxInputSize {
+		return nil, ErrTooLarge
+	}
+
+	if n := CompressBound(len(src)); len(dst) < n {
+		dst = make([]byte, n)
+	}
+
+	e := encoder{src: src, dst: dst, hashTable: make([]uint32, hashTableSize)}
+
+	binary.LittleEndian.PutUint32(dst, uint32(len(src)))
+	e.dpos = 4
+
+	var (
+		step  uint32 = 1
+		limit        = incompressible
+	)
+
+	for {
+		if int(e.pos)+12 >= len(e.src) {
+			e.writeLiterals(uint32(len(e.src))-e.anchor, 0, e.anchor)
+			return e.dst[:e.dpos], nil
+		}
+
+		sequence := uint32(e.src[e.pos+3])<<24 | uint32(e.src[e.pos+2])<<16 | uint32(e.src[e.pos+1])<<8 | uint32(e.src[e.pos+0])
+
+		hash := (sequence * 2654435761) >> hashShift
+		ref := e.hashTable[hash] + uninitHash
+		e.hashTable[hash] = e.pos - uninitHash
+
+		if ((e.pos-ref)>>16) != 0 || uint32(e.src[ref+3])<<24|uint32(e.src[ref+2])<<16|uint32(e.src[ref+1])<<8|uint32(e.src[ref+0]) != sequence {
+			if e.pos-e.anchor > limit {
+				limit <<= 1
+				step += 1 + (step >> 2)
+			}
+			e.pos += step
+			continue
+		}
+
+		if step > 1 {
+			e.hashTable[hash] = ref - uninitHash
+			e.pos -= step - 1
+			step = 1
+			continue
+		}
+		limit = incompressible
+
+		ln := e.pos - e.anchor
+		back := e.pos - ref
+
+		anchor := e.anchor
+
+		e.pos += minMatch
+		ref += minMatch
+		e.anchor = e.pos
+
+		for int(e.pos) < len(e.src)-5 && e.src[e.pos] == e.src[ref] {
+			e.pos++
+			ref++
+		}
+
+		mlLen := e.pos - e.anchor
+
+		e.writeLiterals(ln, mlLen, anchor)
+		e.dst[e.dpos] = uint8(back)
+		e.dst[e.dpos+1] = uint8(back >> 8)
+		e.dpos += 2
+
+		if mlLen > mlMask-1 {
+			mlLen -= mlMask
+			for mlLen > 254 {
+				mlLen -= 255
+
+				e.dst[e.dpos] = 255
+				e.dpos++
+			}
+
+			e.dst[e.dpos] = byte(mlLen)
+			e.dpos++
+		}
+
+		e.anchor = e.pos
+	}
+}

Godeps/_workspace/src/github.com/calmh/logger/LICENSE

@@ -0,0 +1,19 @@
+Copyright (C) 2013 Jakob Borg
+
+Permission is hereby granted, free of charge, to any person obtaining a copy of
+this software and associated documentation files (the "Software"), to deal in
+the Software without restriction, including without limitation the rights to
+use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
+of the Software, and to permit persons to whom the Software is furnished to do
+so, subject to the following conditions:
+
+- The above copyright notice and this permission notice shall be included in
+  all copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

Godeps/_workspace/src/github.com/calmh/logger/README.md

@@ -0,0 +1,15 @@
+logger
+======
+
+A small wrapper around `log` to provide log levels.
+
+Documentation
+-------------
+
+http://godoc.org/github.com/calmh/logger
+
+License
+-------
+
+MIT
+

Godeps/_workspace/src/github.com/calmh/logger/logger.go

@@ -1,6 +1,5 @@
-// Copyright (C) 2014 Jakob Borg and Contributors (see the CONTRIBUTORS file).
-// All rights reserved. Use of this source code is governed by an MIT-style
-// license that can be found in the LICENSE file.
+// Copyright (C) 2014 Jakob Borg. All rights reserved. Use of this source code
+// is governed by an MIT-style license that can be found in the LICENSE file.
 
 // Package logger implements a standardized logger with callback functionality
 package logger
@@ -9,6 +8,7 @@ import (
 	"fmt"
 	"log"
 	"os"
+	"strings"
 	"sync"
 )
 
@@ -23,6 +23,7 @@ const (
 	NumLevels
 )
 
+// A MessageHandler is called with the log level and message text.
 type MessageHandler func(l LogLevel, msg string)
 
 type Logger struct {
@@ -31,6 +32,7 @@ type Logger struct {
 	mut      sync.Mutex
 }
 
+// The default logger logs to standard output with a time prefix.
 var DefaultLogger = New()
 
 func New() *Logger {
@@ -39,26 +41,31 @@ func New() *Logger {
 	}
 }
 
+// AddHandler registers a new MessageHandler to receive messages with the
+// specified log level or above.
 func (l *Logger) AddHandler(level LogLevel, h MessageHandler) {
 	l.mut.Lock()
 	defer l.mut.Unlock()
 	l.handlers[level] = append(l.handlers[level], h)
 }
 
+// See log.SetFlags
 func (l *Logger) SetFlags(flag int) {
 	l.logger.SetFlags(flag)
 }
 
+// See log.SetPrefix
 func (l *Logger) SetPrefix(prefix string) {
 	l.logger.SetPrefix(prefix)
 }
 
 func (l *Logger) callHandlers(level LogLevel, s string) {
 	for _, h := range l.handlers[level] {
-		h(level, s)
+		h(level, strings.TrimSpace(s))
 	}
 }
 
+// Debugln logs a line with a DEBUG prefix.
 func (l *Logger) Debugln(vals ...interface{}) {
 	l.mut.Lock()
 	defer l.mut.Unlock()
@@ -67,6 +74,7 @@ func (l *Logger) Debugln(vals ...interface{}) {
 	l.callHandlers(LevelDebug, s)
 }
 
+// Debugf logs a formatted line with a DEBUG prefix.
 func (l *Logger) Debugf(format string, vals ...interface{}) {
 	l.mut.Lock()
 	defer l.mut.Unlock()
@@ -74,6 +82,8 @@ func (l *Logger) Debugf(format string, vals ...interface{}) {
 	l.logger.Output(2, "DEBUG: "+s)
 	l.callHandlers(LevelDebug, s)
 }
+
+// Infoln logs a line with an INFO prefix.
 func (l *Logger) Infoln(vals ...interface{}) {
 	l.mut.Lock()
 	defer l.mut.Unlock()
@@ -82,6 +92,7 @@ func (l *Logger) Infoln(vals ...interface{}) {
 	l.callHandlers(LevelInfo, s)
 }
 
+// Infof logs a formatted line with an INFO prefix.
 func (l *Logger) Infof(format string, vals ...interface{}) {
 	l.mut.Lock()
 	defer l.mut.Unlock()
@@ -90,6 +101,7 @@ func (l *Logger) Infof(format string, vals ...interface{}) {
 	l.callHandlers(LevelInfo, s)
 }
 
+// Okln logs a line with an OK prefix.
 func (l *Logger) Okln(vals ...interface{}) {
 	l.mut.Lock()
 	defer l.mut.Unlock()
@@ -98,6 +110,7 @@ func (l *Logger) Okln(vals ...interface{}) {
 	l.callHandlers(LevelOK, s)
 }
 
+// Okf logs a formatted line with an OK prefix.
 func (l *Logger) Okf(format string, vals ...interface{}) {
 	l.mut.Lock()
 	defer l.mut.Unlock()
@@ -106,6 +119,7 @@ func (l *Logger) Okf(format string, vals ...interface{}) {
 	l.callHandlers(LevelOK, s)
 }
 
+// Warnln logs a formatted line with a WARNING prefix.
 func (l *Logger) Warnln(vals ...interface{}) {
 	l.mut.Lock()
 	defer l.mut.Unlock()
@@ -114,6 +128,7 @@ func (l *Logger) Warnln(vals ...interface{}) {
 	l.callHandlers(LevelWarn, s)
 }
 
+// Warnf logs a formatted line with a WARNING prefix.
 func (l *Logger) Warnf(format string, vals ...interface{}) {
 	l.mut.Lock()
 	defer l.mut.Unlock()
@@ -122,31 +137,24 @@ func (l *Logger) Warnf(format string, vals ...interface{}) {
 	l.callHandlers(LevelWarn, s)
 }
 
+// Fatalln logs a line with a FATAL prefix and exits the process with exit
+// code 1.
 func (l *Logger) Fatalln(vals ...interface{}) {
 	l.mut.Lock()
 	defer l.mut.Unlock()
 	s := fmt.Sprintln(vals...)
 	l.logger.Output(2, "FATAL: "+s)
 	l.callHandlers(LevelFatal, s)
-	os.Exit(3)
+	os.Exit(1)
 }
 
+// Fatalf logs a formatted line with a FATAL prefix and exits the process with
+// exit code 1.
 func (l *Logger) Fatalf(format string, vals ...interface{}) {
 	l.mut.Lock()
 	defer l.mut.Unlock()
 	s := fmt.Sprintf(format, vals...)
 	l.logger.Output(2, "FATAL: "+s)
 	l.callHandlers(LevelFatal, s)
-	os.Exit(3)
-}
-
-func (l *Logger) FatalErr(err error) {
-	if err != nil {
-		l.mut.Lock()
-		defer l.mut.Unlock()
-		l.logger.SetFlags(l.logger.Flags() | log.Lshortfile)
-		l.logger.Output(2, "FATAL: "+err.Error())
-		l.callHandlers(LevelFatal, err.Error())
-		os.Exit(3)
-	}
+	os.Exit(1)
 }

Godeps/_workspace/src/github.com/calmh/logger/logger_test.go

@@ -0,0 +1,58 @@
+// Copyright (C) 2014 Jakob Borg. All rights reserved. Use of this source code
+// is governed by an MIT-style license that can be found in the LICENSE file.
+
+package logger
+
+import (
+	"strings"
+	"testing"
+)
+
+func TestAPI(t *testing.T) {
+	l := New()
+	l.SetFlags(0)
+	l.SetPrefix("testing")
+
+	debug := 0
+	l.AddHandler(LevelDebug, checkFunc(t, LevelDebug, "test 0", &debug))
+	info := 0
+	l.AddHandler(LevelInfo, checkFunc(t, LevelInfo, "test 1", &info))
+	warn := 0
+	l.AddHandler(LevelWarn, checkFunc(t, LevelWarn, "test 2", &warn))
+	ok := 0
+	l.AddHandler(LevelOK, checkFunc(t, LevelOK, "test 3", &ok))
+
+	l.Debugf("test %d", 0)
+	l.Debugln("test", 0)
+	l.Infof("test %d", 1)
+	l.Infoln("test", 1)
+	l.Warnf("test %d", 2)
+	l.Warnln("test", 2)
+	l.Okf("test %d", 3)
+	l.Okln("test", 3)
+
+	if debug != 2 {
+		t.Errorf("Debug handler called %d != 2 times", debug)
+	}
+	if info != 2 {
+		t.Errorf("Info handler called %d != 2 times", info)
+	}
+	if warn != 2 {
+		t.Errorf("Warn handler called %d != 2 times", warn)
+	}
+	if ok != 2 {
+		t.Errorf("Ok handler called %d != 2 times", ok)
+	}
+}
+
+func checkFunc(t *testing.T, expectl LogLevel, expectmsg string, counter *int) func(LogLevel, string) {
+	return func(l LogLevel, msg string) {
+		*counter++
+		if l != expectl {
+			t.Errorf("Incorrect message level %d != %d", l, expectl)
+		}
+		if !strings.HasSuffix(msg, expectmsg) {
+			t.Errorf("%q does not end with %q", msg, expectmsg)
+		}
+	}
+}

Godeps/_workspace/src/github.com/calmh/osext/osext_procfs.go

@@ -2,12 +2,13 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// +build linux netbsd openbsd
+// +build linux netbsd openbsd solaris
 
 package osext
 
 import (
 	"errors"
+	"fmt"
 	"os"
 	"runtime"
 )
@@ -20,6 +21,8 @@ func executable() (string, error) {
 		return os.Readlink("/proc/curproc/exe")
 	case "openbsd":
 		return os.Readlink("/proc/curproc/file")
+	case "solaris":
+		return os.Readlink(fmt.Sprintf("/proc/%d/path/a.out", os.Getpid()))
 	}
 	return "", errors.New("ExecPath not implemented for " + runtime.GOOS)
 }

Godeps/_workspace/src/github.com/calmh/xdr/.gitignore

@@ -0,0 +1 @@
+coverage.out

Godeps/_workspace/src/github.com/calmh/xdr/.travis.yml

@@ -0,0 +1,19 @@
+language: go
+go:
+- tip
+
+install:
+- export PATH=$PATH:$HOME/gopath/bin
+- go get code.google.com/p/go.tools/cmd/cover
+- go get github.com/mattn/goveralls
+
+script:
+- ./generate.sh
+- go test -coverprofile=coverage.out
+
+after_success:
+- goveralls -coverprofile=coverage.out -service=travis-ci -package=calmh/xdr -repotoken="$COVERALLS_TOKEN"
+
+env:
+  global:
+    secure: SmgnrGfp2zLrA44ChRMpjPeujubt9veZ8Fx/OseMWECmacyV5N/TuDhzIbwo6QwV4xB0sBacoPzvxQbJRVjNKsPiSu72UbcQmQ7flN4Tf7nW09tSh1iW8NgrpBCq/3UYLoBu2iPBEBKm93IK0aGNAKs6oEkB0fU27iTVBwiTXOY=

Godeps/_workspace/src/github.com/calmh/xdr/LICENSE

@@ -1,21 +1,19 @@
-The MIT License (MIT)
-
-Copyright (c) 2011-2014 Twitter, Inc
+Copyright (C) 2014 Jakob Borg.
 
 Permission is hereby granted, free of charge, to any person obtaining a copy
-of this software and associated documentation files (the "Software"), to deal
-in the Software without restriction, including without limitation the rights
-to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
-copies of the Software, and to permit persons to whom the Software is
+of this software and associated documentation files (the "Software"), to
+deal in the Software without restriction, including without limitation the
+rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+sell copies of the Software, and to permit persons to whom the Software is
 furnished to do so, subject to the following conditions:
 
-The above copyright notice and this permission notice shall be included in
-all copies or substantial portions of the Software.
+- The above copyright notice and this permission notice shall be included in
+  all copies or substantial portions of the Software.
 
 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
 AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
-LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
-OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
-THE SOFTWARE.
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+IN THE SOFTWARE.

Godeps/_workspace/src/github.com/calmh/xdr/README.md

@@ -0,0 +1,12 @@
+xdr
+===
+
+[![Build Status](https://img.shields.io/travis/calmh/xdr.svg?style=flat)](https://travis-ci.org/calmh/xdr)
+[![Coverage Status](https://img.shields.io/coveralls/calmh/xdr.svg?style=flat)](https://coveralls.io/r/calmh/xdr?branch=master)
+[![API Documentation](http://img.shields.io/badge/api-Godoc-blue.svg?style=flat)](http://godoc.org/github.com/calmh/xdr)
+[![MIT License](http://img.shields.io/badge/license-MIT-blue.svg?style=flat)](http://opensource.org/licenses/MIT)
+
+This is an XDR encoding/decoding library. It uses code generation and
+not reflection. It supports the IPDR bastardized XDR format when built
+with `-tags ipdr`.
+

Godeps/_workspace/src/github.com/calmh/xdr/bench_test.go

@@ -1,6 +1,5 @@
-// Copyright (C) 2014 Jakob Borg and Contributors (see the CONTRIBUTORS file).
-// All rights reserved. Use of this source code is governed by an MIT-style
-// license that can be found in the LICENSE file.
+// Copyright (C) 2014 Jakob Borg. All rights reserved. Use of this source code
+// is governed by an MIT-style license that can be found in the LICENSE file.
 
 package xdr_test
 
@@ -8,12 +7,15 @@ import (
 	"io"
 	"io/ioutil"
 	"testing"
+
+	"github.com/calmh/xdr"
 )
 
 type XDRBenchStruct struct {
 	I1  uint64
 	I2  uint32
 	I3  uint16
+	I4  uint8
 	Bs0 []byte // max:128
 	Bs1 []byte
 	S0  string // max:128
@@ -25,16 +27,21 @@ var s = XDRBenchStruct{
 	I1:  42,
 	I2:  43,
 	I3:  44,
+	I4:  45,
 	Bs0: []byte{1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18},
 	Bs1: []byte{11, 12, 13, 14, 15, 16, 17, 18, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10},
 	S0:  "Hello World! String one.",
 	S1:  "Hello World! String two.",
 }
-var e = s.MarshalXDR()
+var e []byte
+
+func init() {
+	e, _ = s.MarshalXDR()
+}
 
 func BenchmarkThisMarshal(b *testing.B) {
 	for i := 0; i < b.N; i++ {
-		res = s.MarshalXDR()
+		res, _ = s.MarshalXDR()
 	}
 }
 
@@ -57,6 +64,16 @@ func BenchmarkThisEncode(b *testing.B) {
 	}
 }
 
+func BenchmarkThisEncoder(b *testing.B) {
+	w := xdr.NewWriter(ioutil.Discard)
+	for i := 0; i < b.N; i++ {
+		_, err := s.encodeXDR(w)
+		if err != nil {
+			b.Fatal(err)
+		}
+	}
+}
+
 type repeatReader struct {
 	data []byte
 }
@@ -85,3 +102,16 @@ func BenchmarkThisDecode(b *testing.B) {
 		rr.Reset(e)
 	}
 }
+
+func BenchmarkThisDecoder(b *testing.B) {
+	rr := &repeatReader{e}
+	r := xdr.NewReader(rr)
+	var t XDRBenchStruct
+	for i := 0; i < b.N; i++ {
+		err := t.decodeXDR(r)
+		if err != nil {
+			b.Fatal(err)
+		}
+		rr.Reset(e)
+	}
+}

Godeps/_workspace/src/github.com/calmh/xdr/bench_xdr_test.go

@@ -0,0 +1,199 @@
+// ************************************************************
+// This file is automatically generated by genxdr. Do not edit.
+// ************************************************************
+
+package xdr_test
+
+import (
+	"bytes"
+	"io"
+
+	"github.com/calmh/xdr"
+)
+
+/*
+
+XDRBenchStruct Structure:
+
+ 0                   1                   2                   3
+ 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+|                                                               |
++                         I1 (64 bits)                          +
+|                                                               |
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+|                              I2                               |
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+|            0x0000             |              I3               |
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+|                             uint8                             |
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+|                         Length of Bs0                         |
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+/                                                               /
+\                     Bs0 (variable length)                     \
+/                                                               /
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+|                         Length of Bs1                         |
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+/                                                               /
+\                     Bs1 (variable length)                     \
+/                                                               /
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+|                         Length of S0                          |
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+/                                                               /
+\                     S0 (variable length)                      \
+/                                                               /
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+|                         Length of S1                          |
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+/                                                               /
+\                     S1 (variable length)                      \
+/                                                               /
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+
+
+struct XDRBenchStruct {
+	unsigned hyper I1;
+	unsigned int I2;
+	unsigned int I3;
+	uint8 I4;
+	opaque Bs0<128>;
+	opaque Bs1<>;
+	string S0<128>;
+	string S1<>;
+}
+
+*/
+
+func (o XDRBenchStruct) EncodeXDR(w io.Writer) (int, error) {
+	var xw = xdr.NewWriter(w)
+	return o.encodeXDR(xw)
+}
+
+func (o XDRBenchStruct) MarshalXDR() ([]byte, error) {
+	return o.AppendXDR(make([]byte, 0, 128))
+}
+
+func (o XDRBenchStruct) MustMarshalXDR() []byte {
+	bs, err := o.MarshalXDR()
+	if err != nil {
+		panic(err)
+	}
+	return bs
+}
+
+func (o XDRBenchStruct) AppendXDR(bs []byte) ([]byte, error) {
+	var aw = xdr.AppendWriter(bs)
+	var xw = xdr.NewWriter(&aw)
+	_, err := o.encodeXDR(xw)
+	return []byte(aw), err
+}
+
+func (o XDRBenchStruct) encodeXDR(xw *xdr.Writer) (int, error) {
+	xw.WriteUint64(o.I1)
+	xw.WriteUint32(o.I2)
+	xw.WriteUint16(o.I3)
+	xw.WriteUint8(o.I4)
+	if l := len(o.Bs0); l > 128 {
+		return xw.Tot(), xdr.ElementSizeExceeded("Bs0", l, 128)
+	}
+	xw.WriteBytes(o.Bs0)
+	xw.WriteBytes(o.Bs1)
+	if l := len(o.S0); l > 128 {
+		return xw.Tot(), xdr.ElementSizeExceeded("S0", l, 128)
+	}
+	xw.WriteString(o.S0)
+	xw.WriteString(o.S1)
+	return xw.Tot(), xw.Error()
+}
+
+func (o *XDRBenchStruct) DecodeXDR(r io.Reader) error {
+	xr := xdr.NewReader(r)
+	return o.decodeXDR(xr)
+}
+
+func (o *XDRBenchStruct) UnmarshalXDR(bs []byte) error {
+	var br = bytes.NewReader(bs)
+	var xr = xdr.NewReader(br)
+	return o.decodeXDR(xr)
+}
+
+func (o *XDRBenchStruct) decodeXDR(xr *xdr.Reader) error {
+	o.I1 = xr.ReadUint64()
+	o.I2 = xr.ReadUint32()
+	o.I3 = xr.ReadUint16()
+	o.I4 = xr.ReadUint8()
+	o.Bs0 = xr.ReadBytesMax(128)
+	o.Bs1 = xr.ReadBytes()
+	o.S0 = xr.ReadStringMax(128)
+	o.S1 = xr.ReadString()
+	return xr.Error()
+}
+
+/*
+
+repeatReader Structure:
+
+ 0                   1                   2                   3
+ 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+|                        Length of data                         |
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+/                                                               /
+\                    data (variable length)                     \
+/                                                               /
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+
+
+struct repeatReader {
+	opaque data<>;
+}
+
+*/
+
+func (o repeatReader) EncodeXDR(w io.Writer) (int, error) {
+	var xw = xdr.NewWriter(w)
+	return o.encodeXDR(xw)
+}
+
+func (o repeatReader) MarshalXDR() ([]byte, error) {
+	return o.AppendXDR(make([]byte, 0, 128))
+}
+
+func (o repeatReader) MustMarshalXDR() []byte {
+	bs, err := o.MarshalXDR()
+	if err != nil {
+		panic(err)
+	}
+	return bs
+}
+
+func (o repeatReader) AppendXDR(bs []byte) ([]byte, error) {
+	var aw = xdr.AppendWriter(bs)
+	var xw = xdr.NewWriter(&aw)
+	_, err := o.encodeXDR(xw)
+	return []byte(aw), err
+}
+
+func (o repeatReader) encodeXDR(xw *xdr.Writer) (int, error) {
+	xw.WriteBytes(o.data)
+	return xw.Tot(), xw.Error()
+}
+
+func (o *repeatReader) DecodeXDR(r io.Reader) error {
+	xr := xdr.NewReader(r)
+	return o.decodeXDR(xr)
+}
+
+func (o *repeatReader) UnmarshalXDR(bs []byte) error {
+	var br = bytes.NewReader(bs)
+	var xr = xdr.NewReader(br)
+	return o.decodeXDR(xr)
+}
+
+func (o *repeatReader) decodeXDR(xr *xdr.Reader) error {
+	o.data = xr.ReadBytes()
+	return xr.Error()
+}

Godeps/_workspace/src/github.com/calmh/xdr/cmd/genxdr/main.go

@@ -1,6 +1,5 @@
-// Copyright (C) 2014 Jakob Borg and Contributors (see the CONTRIBUTORS file).
-// All rights reserved. Use of this source code is governed by an MIT-style
-// license that can be found in the LICENSE file.
+// Copyright (C) 2014 Jakob Borg. All rights reserved. Use of this source code
+// is governed by an MIT-style license that can be found in the LICENSE file.
 
 package main
 
@@ -34,11 +33,7 @@ type structInfo struct {
 	Fields []fieldInfo
 }
 
-var headerTpl = template.Must(template.New("header").Parse(`// Copyright (C) 2014 Jakob Borg and Contributors (see the CONTRIBUTORS file).
-// All rights reserved. Use of this source code is governed by an MIT-style
-// license that can be found in the LICENSE file.
-
-// ************************************************************
+var headerTpl = template.Must(template.New("header").Parse(`// ************************************************************
 // This file is automatically generated by genxdr. Do not edit.
 // ************************************************************
 
@@ -48,7 +43,7 @@ import (
 	"bytes"
 	"io"
 
-	"github.com/calmh/syncthing/xdr"
+	"github.com/calmh/xdr"
 )
 `))
 
@@ -58,15 +53,23 @@ func (o {{.TypeName}}) EncodeXDR(w io.Writer) (int, error) {
 	return o.encodeXDR(xw)
 }//+n
 
-func (o {{.TypeName}}) MarshalXDR() []byte {
+func (o {{.TypeName}}) MarshalXDR() ([]byte, error) {
 	return o.AppendXDR(make([]byte, 0, 128))
 }//+n
 
-func (o {{.TypeName}}) AppendXDR(bs []byte) []byte {
+func (o {{.TypeName}}) MustMarshalXDR() []byte {
+	bs, err := o.MarshalXDR()
+	if err != nil {
+		panic(err)
+	}
+	return bs
+}//+n
+
+func (o {{.TypeName}}) AppendXDR(bs []byte) ([]byte, error) {
 	var aw = xdr.AppendWriter(bs)
 	var xw = xdr.NewWriter(&aw)
-	o.encodeXDR(xw)
-	return []byte(aw)
+	_, err := o.encodeXDR(xw)
+	return []byte(aw), err
 }//+n
 
 func (o {{.TypeName}}) encodeXDR(xw *xdr.Writer) (int, error) {
@@ -76,18 +79,21 @@ func (o {{.TypeName}}) encodeXDR(xw *xdr.Writer) (int, error) {
 				xw.Write{{$fieldInfo.Encoder}}({{$fieldInfo.Convert}}(o.{{$fieldInfo.Name}}))
 			{{else if $fieldInfo.IsBasic}}
 				{{if ge $fieldInfo.Max 1}}
-					if len(o.{{$fieldInfo.Name}}) > {{$fieldInfo.Max}} {
-						return xw.Tot(), xdr.ErrElementSizeExceeded
+					if l := len(o.{{$fieldInfo.Name}}); l > {{$fieldInfo.Max}} {
+						return xw.Tot(), xdr.ElementSizeExceeded("{{$fieldInfo.Name}}", l, {{$fieldInfo.Max}})
 					}
 				{{end}}
 				xw.Write{{$fieldInfo.Encoder}}(o.{{$fieldInfo.Name}})
 			{{else}}
-				o.{{$fieldInfo.Name}}.encodeXDR(xw)
+				_, err := o.{{$fieldInfo.Name}}.encodeXDR(xw)
+				if err != nil {
+					return xw.Tot(), err
+				}
 			{{end}}
 		{{else}}
 			{{if ge $fieldInfo.Max 1}}
-				if len(o.{{$fieldInfo.Name}}) > {{$fieldInfo.Max}} {
-					return xw.Tot(), xdr.ErrElementSizeExceeded
+				if l := len(o.{{$fieldInfo.Name}}); l > {{$fieldInfo.Max}} {
+					return xw.Tot(), xdr.ElementSizeExceeded("{{$fieldInfo.Name}}", l, {{$fieldInfo.Max}})
 				}
 			{{end}}
 			xw.WriteUint32(uint32(len(o.{{$fieldInfo.Name}})))
@@ -97,7 +103,10 @@ func (o {{.TypeName}}) encodeXDR(xw *xdr.Writer) (int, error) {
 			{{else if $fieldInfo.IsBasic}}
 				xw.Write{{$fieldInfo.Encoder}}(o.{{$fieldInfo.Name}}[i])
 			{{else}}
-				o.{{$fieldInfo.Name}}[i].encodeXDR(xw)
+				_, err := o.{{$fieldInfo.Name}}[i].encodeXDR(xw)
+				if err != nil {
+					return xw.Tot(), err
+				}
 			{{end}}
 			}
 		{{end}}
@@ -134,7 +143,7 @@ func (o *{{.TypeName}}) decodeXDR(xr *xdr.Reader) error {
 			_{{$fieldInfo.Name}}Size := int(xr.ReadUint32())
 			{{if ge $fieldInfo.Max 1}}
 				if _{{$fieldInfo.Name}}Size > {{$fieldInfo.Max}} {
-					return xdr.ErrElementSizeExceeded
+					return xdr.ElementSizeExceeded("{{$fieldInfo.Name}}", _{{$fieldInfo.Name}}Size, {{$fieldInfo.Max}})
 				}
 			{{end}}
 			o.{{$fieldInfo.Name}} = make([]{{$fieldInfo.FieldType}}, _{{$fieldInfo.Name}}Size)
@@ -160,6 +169,8 @@ type typeSet struct {
 }
 
 var xdrEncoders = map[string]typeSet{
+	"int8":   typeSet{"uint8", "Uint8"},
+	"uint8":  typeSet{"", "Uint8"},
 	"int16":  typeSet{"uint16", "Uint16"},
 	"uint16": typeSet{"", "Uint16"},
 	"int32":  typeSet{"uint32", "Uint32"},

Godeps/_workspace/src/github.com/calmh/xdr/debug.go

@@ -0,0 +1,16 @@
+// Copyright (C) 2014 Jakob Borg. All rights reserved. Use of this source code
+// is governed by an MIT-style license that can be found in the LICENSE file.
+
+package xdr
+
+import (
+	"log"
+	"os"
+)
+
+var (
+	debug = len(os.Getenv("XDRTRACE")) > 0
+	dl    = log.New(os.Stdout, "xdr: ", log.Lshortfile|log.Ltime|log.Lmicroseconds)
+)
+
+const maxDebugBytes = 32

Godeps/_workspace/src/github.com/calmh/xdr/doc.go

@@ -0,0 +1,5 @@
+// Copyright (C) 2014 Jakob Borg. All rights reserved. Use of this source code
+// is governed by an MIT-style license that can be found in the LICENSE file.
+
+// Package xdr implements an XDR (RFC 4506) encoder/decoder.
+package xdr

Godeps/_workspace/src/github.com/calmh/xdr/encdec_test.go

@@ -0,0 +1,79 @@
+// Copyright (C) 2014 Jakob Borg. All rights reserved. Use of this source code
+// is governed by an MIT-style license that can be found in the LICENSE file.
+
+package xdr_test
+
+import (
+	"bytes"
+	"math/rand"
+	"reflect"
+	"testing"
+	"testing/quick"
+
+	"github.com/calmh/xdr"
+)
+
+// Contains all supported types
+type TestStruct struct {
+	I    int
+	I8   int8
+	UI8  uint8
+	I16  int16
+	UI16 uint16
+	I32  int32
+	UI32 uint32
+	I64  int64
+	UI64 uint64
+	BS   []byte // max:1024
+	S    string // max:1024
+	C    Opaque
+	SS   []string // max:1024
+}
+
+type Opaque [32]byte
+
+func (u *Opaque) encodeXDR(w *xdr.Writer) (int, error) {
+	return w.WriteRaw(u[:])
+}
+
+func (u *Opaque) decodeXDR(r *xdr.Reader) (int, error) {
+	return r.ReadRaw(u[:])
+}
+
+func (Opaque) Generate(rand *rand.Rand, size int) reflect.Value {
+	var u Opaque
+	for i := range u[:] {
+		u[i] = byte(rand.Int())
+	}
+	return reflect.ValueOf(u)
+}
+
+func TestEncDec(t *testing.T) {
+	fn := func(t0 TestStruct) bool {
+		bs, err := t0.MarshalXDR()
+		if err != nil {
+			t.Fatal(err)
+		}
+		var t1 TestStruct
+		err = t1.UnmarshalXDR(bs)
+		if err != nil {
+			t.Fatal(err)
+		}
+
+		// Not comparing with DeepEqual since we'll unmarshal nil slices as empty
+		if t0.I != t1.I ||
+			t0.I16 != t1.I16 || t0.UI16 != t1.UI16 ||
+			t0.I32 != t1.I32 || t0.UI32 != t1.UI32 ||
+			t0.I64 != t1.I64 || t0.UI64 != t1.UI64 ||
+			bytes.Compare(t0.BS, t1.BS) != 0 ||
+			t0.S != t1.S || t0.C != t1.C {
+			t.Logf("%#v", t0)
+			t.Logf("%#v", t1)
+			return false
+		}
+		return true
+	}
+	if err := quick.Check(fn, nil); err != nil {
+		t.Error(err)
+	}
+}

Godeps/_workspace/src/github.com/calmh/xdr/encdec_xdr_test.go

@@ -0,0 +1,174 @@
+// ************************************************************
+// This file is automatically generated by genxdr. Do not edit.
+// ************************************************************
+
+package xdr_test
+
+import (
+	"bytes"
+	"io"
+
+	"github.com/calmh/xdr"
+)
+
+/*
+
+TestStruct Structure:
+
+ 0                   1                   2                   3
+ 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+|                              int                              |
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+|                             int8                              |
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+|                             uint8                             |
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+|                             int16                             |
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+|            0x0000             |             UI16              |
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+|                             int32                             |
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+|                             UI32                              |
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+|                                                               |
++                         I64 (64 bits)                         +
+|                                                               |
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+|                                                               |
++                        UI64 (64 bits)                         +
+|                                                               |
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+|                         Length of BS                          |
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+/                                                               /
+\                     BS (variable length)                      \
+/                                                               /
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+|                          Length of S                          |
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+/                                                               /
+\                      S (variable length)                      \
+/                                                               /
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+|                            Opaque                             |
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+|                         Number of SS                          |
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+|                         Length of SS                          |
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+/                                                               /
+\                     SS (variable length)                      \
+/                                                               /
++-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+
+
+struct TestStruct {
+	int I;
+	int8 I8;
+	uint8 UI8;
+	int16 I16;
+	unsigned int UI16;
+	int32 I32;
+	unsigned int UI32;
+	hyper I64;
+	unsigned hyper UI64;
+	opaque BS<1024>;
+	string S<1024>;
+	Opaque C;
+	string SS<1024>;
+}
+
+*/
+
+func (o TestStruct) EncodeXDR(w io.Writer) (int, error) {
+	var xw = xdr.NewWriter(w)
+	return o.encodeXDR(xw)
+}
+
+func (o TestStruct) MarshalXDR() ([]byte, error) {
+	return o.AppendXDR(make([]byte, 0, 128))
+}
+
+func (o TestStruct) MustMarshalXDR() []byte {
+	bs, err := o.MarshalXDR()
+	if err != nil {
+		panic(err)
+	}
+	return bs
+}
+
+func (o TestStruct) AppendXDR(bs []byte) ([]byte, error) {
+	var aw = xdr.AppendWriter(bs)
+	var xw = xdr.NewWriter(&aw)
+	_, err := o.encodeXDR(xw)
+	return []byte(aw), err
+}
+
+func (o TestStruct) encodeXDR(xw *xdr.Writer) (int, error) {
+	xw.WriteUint64(uint64(o.I))
+	xw.WriteUint8(uint8(o.I8))
+	xw.WriteUint8(o.UI8)
+	xw.WriteUint16(uint16(o.I16))
+	xw.WriteUint16(o.UI16)
+	xw.WriteUint32(uint32(o.I32))
+	xw.WriteUint32(o.UI32)
+	xw.WriteUint64(uint64(o.I64))
+	xw.WriteUint64(o.UI64)
+	if l := len(o.BS); l > 1024 {
+		return xw.Tot(), xdr.ElementSizeExceeded("BS", l, 1024)
+	}
+	xw.WriteBytes(o.BS)
+	if l := len(o.S); l > 1024 {
+		return xw.Tot(), xdr.ElementSizeExceeded("S", l, 1024)
+	}
+	xw.WriteString(o.S)
+	_, err := o.C.encodeXDR(xw)
+	if err != nil {
+		return xw.Tot(), err
+	}
+	if l := len(o.SS); l > 1024 {
+		return xw.Tot(), xdr.ElementSizeExceeded("SS", l, 1024)
+	}
+	xw.WriteUint32(uint32(len(o.SS)))
+	for i := range o.SS {
+		xw.WriteString(o.SS[i])
+	}
+	return xw.Tot(), xw.Error()
+}
+
+func (o *TestStruct) DecodeXDR(r io.Reader) error {
+	xr := xdr.NewReader(r)
+	return o.decodeXDR(xr)
+}
+
+func (o *TestStruct) UnmarshalXDR(bs []byte) error {
+	var br = bytes.NewReader(bs)
+	var xr = xdr.NewReader(br)
+	return o.decodeXDR(xr)
+}
+
+func (o *TestStruct) decodeXDR(xr *xdr.Reader) error {
+	o.I = int(xr.ReadUint64())
+	o.I8 = int8(xr.ReadUint8())
+	o.UI8 = xr.ReadUint8()
+	o.I16 = int16(xr.ReadUint16())
+	o.UI16 = xr.ReadUint16()
+	o.I32 = int32(xr.ReadUint32())
+	o.UI32 = xr.ReadUint32()
+	o.I64 = int64(xr.ReadUint64())
+	o.UI64 = xr.ReadUint64()
+	o.BS = xr.ReadBytesMax(1024)
+	o.S = xr.ReadStringMax(1024)
+	(&o.C).decodeXDR(xr)
+	_SSSize := int(xr.ReadUint32())
+	if _SSSize > 1024 {
+		return xdr.ElementSizeExceeded("SS", _SSSize, 1024)
+	}
+	o.SS = make([]string, _SSSize)
+	for i := range o.SS {
+		o.SS[i] = xr.ReadString()
+	}
+	return xr.Error()
+}

Godeps/_workspace/src/github.com/calmh/xdr/generate.sh

@@ -0,0 +1,4 @@
+#!/bin/sh
+
+go run cmd/genxdr/main.go -- bench_test.go > bench_xdr_test.go
+go run cmd/genxdr/main.go -- encdec_test.go > encdec_xdr_test.go

Godeps/_workspace/src/github.com/calmh/xdr/pad_ipdr.go

@@ -0,0 +1,10 @@
+// Copyright (C) 2014 Jakob Borg. All rights reserved. Use of this source code
+// is governed by an MIT-style license that can be found in the LICENSE file.
+
+// +build ipdr
+
+package xdr
+
+func pad(l int) int {
+	return 0
+}

Godeps/_workspace/src/github.com/calmh/xdr/pad_xdr.go

@@ -0,0 +1,14 @@
+// Copyright (C) 2014 Jakob Borg. All rights reserved. Use of this source code
+// is governed by an MIT-style license that can be found in the LICENSE file.
+
+// +build !ipdr
+
+package xdr
+
+func pad(l int) int {
+	d := l % 4
+	if d == 0 {
+		return 0
+	}
+	return 4 - d
+}

Godeps/_workspace/src/github.com/calmh/xdr/reader.go

@@ -5,20 +5,16 @@
 package xdr
 
 import (
-	"errors"
+	"fmt"
 	"io"
-	"time"
+	"reflect"
+	"unsafe"
 )
 
-var ErrElementSizeExceeded = errors.New("element size exceeded")
-
 type Reader struct {
-	r    io.Reader
-	tot  int
-	err  error
-	b    [8]byte
-	sb   []byte
-	last time.Time
+	r   io.Reader
+	err error
+	b   [8]byte
 }
 
 func NewReader(r io.Reader) *Reader {
@@ -27,24 +23,28 @@ func NewReader(r io.Reader) *Reader {
 	}
 }
 
-func (r *Reader) ReadString() string {
-	if r.sb == nil {
-		r.sb = make([]byte, 64)
-	} else {
-		r.sb = r.sb[:cap(r.sb)]
+func (r *Reader) ReadRaw(bs []byte) (int, error) {
+	if r.err != nil {
+		return 0, r.err
 	}
-	r.sb = r.ReadBytesInto(r.sb)
-	return string(r.sb)
+
+	var n int
+	n, r.err = io.ReadFull(r.r, bs)
+	return n, r.err
+}
+
+func (r *Reader) ReadString() string {
+	return r.ReadStringMax(0)
 }
 
 func (r *Reader) ReadStringMax(max int) string {
-	if r.sb == nil {
-		r.sb = make([]byte, 64)
-	} else {
-		r.sb = r.sb[:cap(r.sb)]
+	buf := r.ReadBytesMaxInto(max, nil)
+	bh := (*reflect.SliceHeader)(unsafe.Pointer(&buf))
+	sh := reflect.StringHeader{
+		Data: bh.Data,
+		Len:  bh.Len,
 	}
-	r.sb = r.ReadBytesMaxInto(max, r.sb)
-	return string(r.sb)
+	return *((*string)(unsafe.Pointer(&sh)))
 }
 
 func (r *Reader) ReadBytes() []byte {
@@ -63,65 +63,54 @@ func (r *Reader) ReadBytesMaxInto(max int, dst []byte) []byte {
 	if r.err != nil {
 		return nil
 	}
-	r.last = time.Now()
-	s := r.tot
 
 	l := int(r.ReadUint32())
 	if r.err != nil {
 		return nil
 	}
 	if max > 0 && l > max {
-		r.err = ErrElementSizeExceeded
+		r.err = ElementSizeExceeded("bytes field", l, max)
 		return nil
 	}
 
-	if l+pad(l) > len(dst) {
-		dst = make([]byte, l+pad(l))
+	if fullLen := l + pad(l); fullLen > len(dst) {
+		dst = make([]byte, fullLen)
 	} else {
-		dst = dst[:l+pad(l)]
+		dst = dst[:fullLen]
 	}
 
 	var n int
 	n, r.err = io.ReadFull(r.r, dst)
 	if r.err != nil {
 		if debug {
-			dl.Debugf("@0x%x: rd bytes (%d): %v", s, len(dst), r.err)
+			dl.Printf("rd bytes (%d): %v", len(dst), r.err)
 		}
 		return nil
 	}
-	r.tot += n
 
 	if debug {
 		if n > maxDebugBytes {
-			dl.Debugf("@0x%x: rd bytes (%d): %x...", s, len(dst), dst[:maxDebugBytes])
+			dl.Printf("rd bytes (%d): %x...", len(dst), dst[:maxDebugBytes])
 		} else {
-			dl.Debugf("@0x%x: rd bytes (%d): %x", s, len(dst), dst)
+			dl.Printf("rd bytes (%d): %x", len(dst), dst)
 		}
 	}
 	return dst[:l]
 }
 
 func (r *Reader) ReadBool() bool {
-	return r.ReadUint32() != 0
-}
-
-func (r *Reader) ReadUint16() uint16 {
-	return uint16(r.ReadUint32())
+	return r.ReadUint8() != 0
 }
 
 func (r *Reader) ReadUint32() uint32 {
 	if r.err != nil {
 		return 0
 	}
-	r.last = time.Now()
-	s := r.tot
 
-	var n int
-	n, r.err = io.ReadFull(r.r, r.b[:4])
-	r.tot += n
+	_, r.err = io.ReadFull(r.r, r.b[:4])
 	if r.err != nil {
 		if debug {
-			dl.Debugf("@0x%x: rd uint32: %v", r.tot, r.err)
+			dl.Printf("rd uint32: %v", r.err)
 		}
 		return 0
 	}
@@ -129,7 +118,7 @@ func (r *Reader) ReadUint32() uint32 {
 	v := uint32(r.b[3]) | uint32(r.b[2])<<8 | uint32(r.b[1])<<16 | uint32(r.b[0])<<24
 
 	if debug {
-		dl.Debugf("@0x%x: rd uint32=%d (0x%08x)", s, v, v)
+		dl.Printf("rd uint32=%d (0x%08x)", v, v)
 	}
 	return v
 }
@@ -138,15 +127,11 @@ func (r *Reader) ReadUint64() uint64 {
 	if r.err != nil {
 		return 0
 	}
-	r.last = time.Now()
-	s := r.tot
 
-	var n int
-	n, r.err = io.ReadFull(r.r, r.b[:8])
-	r.tot += n
+	_, r.err = io.ReadFull(r.r, r.b[:8])
 	if r.err != nil {
 		if debug {
-			dl.Debugf("@0x%x: rd uint64: %v", r.tot, r.err)
+			dl.Printf("rd uint64: %v", r.err)
 		}
 		return 0
 	}
@@ -155,19 +140,27 @@ func (r *Reader) ReadUint64() uint64 {
 		uint64(r.b[3])<<32 | uint64(r.b[2])<<40 | uint64(r.b[1])<<48 | uint64(r.b[0])<<56
 
 	if debug {
-		dl.Debugf("@0x%x: rd uint64=%d (0x%016x)", s, v, v)
+		dl.Printf("rd uint64=%d (0x%016x)", v, v)
 	}
 	return v
 }
 
-func (r *Reader) Tot() int {
-	return r.tot
+type XDRError struct {
+	op  string
+	err error
+}
+
+func (e XDRError) Error() string {
+	return "xdr " + e.op + ": " + e.err.Error()
 }
 
 func (r *Reader) Error() error {
-	return r.err
+	if r.err == nil {
+		return nil
+	}
+	return XDRError{"read", r.err}
 }
 
-func (r *Reader) LastRead() time.Time {
-	return r.last
+func ElementSizeExceeded(field string, size, limit int) error {
+	return fmt.Errorf("%s exceeds size limit; %d > %d", field, size, limit)
 }

Godeps/_workspace/src/github.com/calmh/xdr/reader_ipdr.go

@@ -0,0 +1,49 @@
+// Copyright (C) 2014 Jakob Borg and Contributors (see the CONTRIBUTORS file).
+// All rights reserved. Use of this source code is governed by an MIT-style
+// license that can be found in the LICENSE file.
+
+// +build ipdr
+
+package xdr
+
+import "io"
+
+func (r *Reader) ReadUint8() uint8 {
+	if r.err != nil {
+		return 0
+	}
+
+	_, r.err = io.ReadFull(r.r, r.b[:1])
+	if r.err != nil {
+		if debug {
+			dl.Printf("rd uint8: %v", r.err)
+		}
+		return 0
+	}
+
+	if debug {
+		dl.Printf("rd uint8=%d (0x%02x)", r.b[0], r.b[0])
+	}
+	return r.b[0]
+}
+
+func (r *Reader) ReadUint16() uint16 {
+	if r.err != nil {
+		return 0
+	}
+
+	_, r.err = io.ReadFull(r.r, r.b[:2])
+	if r.err != nil {
+		if debug {
+			dl.Printf("rd uint16: %v", r.err)
+		}
+		return 0
+	}
+
+	v := uint16(r.b[1]) | uint16(r.b[0])<<8
+
+	if debug {
+		dl.Printf("rd uint16=%d (0x%04x)", v, v)
+	}
+	return v
+}

Godeps/_workspace/src/github.com/calmh/xdr/reader_xdr.go

@@ -2,5 +2,14 @@
 // All rights reserved. Use of this source code is governed by an MIT-style
 // license that can be found in the LICENSE file.
 
-// Package xdr implements an XDR (RFC 4506) encoder/decoder.
+// +build !ipdr
+
 package xdr
+
+func (r *Reader) ReadUint8() uint8 {
+	return uint8(r.ReadUint32())
+}
+
+func (r *Reader) ReadUint16() uint16 {
+	return uint16(r.ReadUint32())
+}

Godeps/_workspace/src/github.com/calmh/xdr/refl_test.go

@@ -1,6 +1,5 @@
-// Copyright (C) 2014 Jakob Borg and Contributors (see the CONTRIBUTORS file).
-// All rights reserved. Use of this source code is governed by an MIT-style
-// license that can be found in the LICENSE file.
+// Copyright (C) 2014 Jakob Borg. All rights reserved. Use of this source code
+// is governed by an MIT-style license that can be found in the LICENSE file.
 
 // +build refl
 

Godeps/_workspace/src/github.com/calmh/xdr/writer.go

@@ -1,30 +1,21 @@
-// Copyright (C) 2014 Jakob Borg and Contributors (see the CONTRIBUTORS file).
-// All rights reserved. Use of this source code is governed by an MIT-style
-// license that can be found in the LICENSE file.
+// Copyright (C) 2014 Jakob Borg. All rights reserved. Use of this source code
+// is governed by an MIT-style license that can be found in the LICENSE file.
 
 package xdr
 
 import (
 	"io"
-	"time"
+	"reflect"
+	"unsafe"
 )
 
-func pad(l int) int {
-	d := l % 4
-	if d == 0 {
-		return 0
-	}
-	return 4 - d
-}
-
 var padBytes = []byte{0, 0, 0}
 
 type Writer struct {
-	w    io.Writer
-	tot  int
-	err  error
-	b    [8]byte
-	last time.Time
+	w   io.Writer
+	tot int
+	err error
+	b   [8]byte
 }
 
 type AppendWriter []byte
@@ -40,8 +31,24 @@ func NewWriter(w io.Writer) *Writer {
 	}
 }
 
+func (w *Writer) WriteRaw(bs []byte) (int, error) {
+	if w.err != nil {
+		return 0, w.err
+	}
+
+	var n int
+	n, w.err = w.w.Write(bs)
+	return n, w.err
+}
+
 func (w *Writer) WriteString(s string) (int, error) {
-	return w.WriteBytes([]byte(s))
+	sh := *((*reflect.StringHeader)(unsafe.Pointer(&s)))
+	bh := reflect.SliceHeader{
+		Data: sh.Data,
+		Len:  sh.Len,
+		Cap:  sh.Len,
+	}
+	return w.WriteBytes(*(*[]byte)(unsafe.Pointer(&bh)))
 }
 
 func (w *Writer) WriteBytes(bs []byte) (int, error) {
@@ -49,7 +56,6 @@ func (w *Writer) WriteBytes(bs []byte) (int, error) {
 		return 0, w.err
 	}
 
-	w.last = time.Now()
 	w.WriteUint32(uint32(len(bs)))
 	if w.err != nil {
 		return 0, w.err
@@ -57,9 +63,9 @@ func (w *Writer) WriteBytes(bs []byte) (int, error) {
 
 	if debug {
 		if len(bs) > maxDebugBytes {
-			dl.Debugf("wr bytes (%d): %x...", len(bs), bs[:maxDebugBytes])
+			dl.Printf("wr bytes (%d): %x...", len(bs), bs[:maxDebugBytes])
 		} else {
-			dl.Debugf("wr bytes (%d): %x", len(bs), bs)
+			dl.Printf("wr bytes (%d): %x", len(bs), bs)
 		}
 	}
 
@@ -78,24 +84,19 @@ func (w *Writer) WriteBytes(bs []byte) (int, error) {
 
 func (w *Writer) WriteBool(v bool) (int, error) {
 	if v {
-		return w.WriteUint32(1)
+		return w.WriteUint8(1)
 	} else {
-		return w.WriteUint32(0)
+		return w.WriteUint8(0)
 	}
 }
 
-func (w *Writer) WriteUint16(v uint16) (int, error) {
-	return w.WriteUint32(uint32(v))
-}
-
 func (w *Writer) WriteUint32(v uint32) (int, error) {
 	if w.err != nil {
 		return 0, w.err
 	}
 
-	w.last = time.Now()
 	if debug {
-		dl.Debugf("wr uint32=%d", v)
+		dl.Printf("wr uint32=%d", v)
 	}
 
 	w.b[0] = byte(v >> 24)
@@ -114,9 +115,8 @@ func (w *Writer) WriteUint64(v uint64) (int, error) {
 		return 0, w.err
 	}
 
-	w.last = time.Now()
 	if debug {
-		dl.Debugf("wr uint64=%d", v)
+		dl.Printf("wr uint64=%d", v)
 	}
 
 	w.b[0] = byte(v >> 56)
@@ -139,9 +139,8 @@ func (w *Writer) Tot() int {
 }
 
 func (w *Writer) Error() error {
-	return w.err
-}
-
-func (w *Writer) LastWrite() time.Time {
-	return w.last
+	if w.err == nil {
+		return nil
+	}
+	return XDRError{"write", w.err}
 }

Godeps/_workspace/src/github.com/calmh/xdr/writer_ipdr.go

@@ -0,0 +1,41 @@
+// Copyright (C) 2014 Jakob Borg. All rights reserved. Use of this source code
+// is governed by an MIT-style license that can be found in the LICENSE file.
+
+// +build ipdr
+
+package xdr
+
+func (w *Writer) WriteUint8(v uint8) (int, error) {
+	if w.err != nil {
+		return 0, w.err
+	}
+
+	if debug {
+		dl.Printf("wr uint8=%d", v)
+	}
+
+	w.b[0] = byte(v)
+
+	var l int
+	l, w.err = w.w.Write(w.b[:1])
+	w.tot += l
+	return l, w.err
+}
+
+func (w *Writer) WriteUint16(v uint16) (int, error) {
+	if w.err != nil {
+		return 0, w.err
+	}
+
+	if debug {
+		dl.Printf("wr uint8=%d", v)
+	}
+
+	w.b[0] = byte(v >> 8)
+	w.b[1] = byte(v)
+
+	var l int
+	l, w.err = w.w.Write(w.b[:2])
+	w.tot += l
+	return l, w.err
+}

Godeps/_workspace/src/github.com/calmh/xdr/writer_xdr.go

@@ -0,0 +1,14 @@
+// Copyright (C) 2014 Jakob Borg. All rights reserved. Use of this source code
+// is governed by an MIT-style license that can be found in the LICENSE file.
+
+// +build !ipdr
+
+package xdr
+
+func (w *Writer) WriteUint8(v uint8) (int, error) {
+	return w.WriteUint32(uint32(v))
+}
+
+func (w *Writer) WriteUint16(v uint16) (int, error) {
+	return w.WriteUint32(uint32(v))
+}

Godeps/_workspace/src/github.com/calmh/xdr/xdr_test.go

@@ -1,32 +1,15 @@
-// Copyright (C) 2014 Jakob Borg and Contributors (see the CONTRIBUTORS file).
-// All rights reserved. Use of this source code is governed by an MIT-style
-// license that can be found in the LICENSE file.
+// Copyright (C) 2014 Jakob Borg. All rights reserved. Use of this source code
+// is governed by an MIT-style license that can be found in the LICENSE file.
 
 package xdr
 
 import (
 	"bytes"
+	"strings"
 	"testing"
 	"testing/quick"
 )
 
-func TestPad(t *testing.T) {
-	tests := [][]int{
-		{0, 0},
-		{1, 3},
-		{2, 2},
-		{3, 1},
-		{4, 0},
-		{32, 0},
-		{33, 3},
-	}
-	for _, tc := range tests {
-		if p := pad(tc[0]); p != tc[1] {
-			t.Errorf("Incorrect padding for %d bytes, %d != %d", tc[0], p, tc[1])
-		}
-	}
-}
-
 func TestBytesNil(t *testing.T) {
 	fn := func(bs []byte) bool {
 		var b = new(bytes.Buffer)
@@ -78,14 +61,14 @@ func TestReadBytesMaxInto(t *testing.T) {
 				if read := len(bs); read != tot {
 					t.Errorf("Incorrect read bytes, wrote=%d, buf=%d, max=%d, read=%d", tot, tot+diff, max, read)
 				}
-			} else if r.err != ErrElementSizeExceeded {
+			} else if !strings.Contains(r.err.Error(), "exceeds size") {
 				t.Errorf("Unexpected non-ErrElementSizeExceeded error for wrote=%d, max=%d: %v", tot, max, r.err)
 			}
 		}
 	}
 }
 
-func TestReadBytesMaxIntoNil(t *testing.T) {
+func TestReadStringMax(t *testing.T) {
 	for tot := 42; tot < 72; tot++ {
 		for max := 0; max < 128; max++ {
 			var b = new(bytes.Buffer)
@@ -95,14 +78,14 @@ func TestReadBytesMaxIntoNil(t *testing.T) {
 			var toWrite = make([]byte, tot)
 			w.WriteBytes(toWrite)
 
-			var bs = r.ReadBytesMaxInto(max, nil)
-			var read = len(bs)
+			var str = r.ReadStringMax(max)
+			var read = len(str)
 
 			if max == 0 || tot <= max {
 				if read != tot {
 					t.Errorf("Incorrect read bytes, wrote=%d, max=%d, read=%d", tot, max, read)
 				}
-			} else if r.err != ErrElementSizeExceeded {
+			} else if !strings.Contains(r.err.Error(), "exceeds size") {
 				t.Errorf("Unexpected non-ErrElementSizeExceeded error for wrote=%d, max=%d, read=%d: %v", tot, max, read, r.err)
 			}
 		}

Godeps/_workspace/src/github.com/golang/groupcache/lru/lru.go

@@ -1,121 +0,0 @@
-/*
-Copyright 2013 Google Inc.
-
-Licensed under the Apache License, Version 2.0 (the "License");
-you may not use this file except in compliance with the License.
-You may obtain a copy of the License at
-
-     http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software
-distributed under the License is distributed on an "AS IS" BASIS,
-WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-See the License for the specific language governing permissions and
-limitations under the License.
-*/
-
-// Package lru implements an LRU cache.
-package lru
-
-import "container/list"
-
-// Cache is an LRU cache. It is not safe for concurrent access.
-type Cache struct {
-	// MaxEntries is the maximum number of cache entries before
-	// an item is evicted. Zero means no limit.
-	MaxEntries int
-
-	// OnEvicted optionally specificies a callback function to be
-	// executed when an entry is purged from the cache.
-	OnEvicted func(key Key, value interface{})
-
-	ll    *list.List
-	cache map[interface{}]*list.Element
-}
-
-// A Key may be any value that is comparable. See http://golang.org/ref/spec#Comparison_operators
-type Key interface{}
-
-type entry struct {
-	key   Key
-	value interface{}
-}
-
-// New creates a new Cache.
-// If maxEntries is zero, the cache has no limit and it's assumed
-// that eviction is done by the caller.
-func New(maxEntries int) *Cache {
-	return &Cache{
-		MaxEntries: maxEntries,
-		ll:         list.New(),
-		cache:      make(map[interface{}]*list.Element),
-	}
-}
-
-// Add adds a value to the cache.
-func (c *Cache) Add(key Key, value interface{}) {
-	if c.cache == nil {
-		c.cache = make(map[interface{}]*list.Element)
-		c.ll = list.New()
-	}
-	if ee, ok := c.cache[key]; ok {
-		c.ll.MoveToFront(ee)
-		ee.Value.(*entry).value = value
-		return
-	}
-	ele := c.ll.PushFront(&entry{key, value})
-	c.cache[key] = ele
-	if c.MaxEntries != 0 && c.ll.Len() > c.MaxEntries {
-		c.RemoveOldest()
-	}
-}
-
-// Get looks up a key's value from the cache.
-func (c *Cache) Get(key Key) (value interface{}, ok bool) {
-	if c.cache == nil {
-		return
-	}
-	if ele, hit := c.cache[key]; hit {
-		c.ll.MoveToFront(ele)
-		return ele.Value.(*entry).value, true
-	}
-	return
-}
-
-// Remove removes the provided key from the cache.
-func (c *Cache) Remove(key Key) {
-	if c.cache == nil {
-		return
-	}
-	if ele, hit := c.cache[key]; hit {
-		c.removeElement(ele)
-	}
-}
-
-// RemoveOldest removes the oldest item from the cache.
-func (c *Cache) RemoveOldest() {
-	if c.cache == nil {
-		return
-	}
-	ele := c.ll.Back()
-	if ele != nil {
-		c.removeElement(ele)
-	}
-}
-
-func (c *Cache) removeElement(e *list.Element) {
-	c.ll.Remove(e)
-	kv := e.Value.(*entry)
-	delete(c.cache, kv.key)
-	if c.OnEvicted != nil {
-		c.OnEvicted(kv.key, kv.value)
-	}
-}
-
-// Len returns the number of items in the cache.
-func (c *Cache) Len() int {
-	if c.cache == nil {
-		return 0
-	}
-	return c.ll.Len()
-}

Godeps/_workspace/src/github.com/golang/groupcache/lru/lru_test.go

@@ -1,73 +0,0 @@
-/*
-Copyright 2013 Google Inc.
-
-Licensed under the Apache License, Version 2.0 (the "License");
-you may not use this file except in compliance with the License.
-You may obtain a copy of the License at
-
-     http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software
-distributed under the License is distributed on an "AS IS" BASIS,
-WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-See the License for the specific language governing permissions and
-limitations under the License.
-*/
-
-package lru
-
-import (
-	"testing"
-)
-
-type simpleStruct struct {
-	int
-	string
-}
-
-type complexStruct struct {
-	int
-	simpleStruct
-}
-
-var getTests = []struct {
-	name       string
-	keyToAdd   interface{}
-	keyToGet   interface{}
-	expectedOk bool
-}{
-	{"string_hit", "myKey", "myKey", true},
-	{"string_miss", "myKey", "nonsense", false},
-	{"simple_struct_hit", simpleStruct{1, "two"}, simpleStruct{1, "two"}, true},
-	{"simeple_struct_miss", simpleStruct{1, "two"}, simpleStruct{0, "noway"}, false},
-	{"complex_struct_hit", complexStruct{1, simpleStruct{2, "three"}},
-		complexStruct{1, simpleStruct{2, "three"}}, true},
-}
-
-func TestGet(t *testing.T) {
-	for _, tt := range getTests {
-		lru := New(0)
-		lru.Add(tt.keyToAdd, 1234)
-		val, ok := lru.Get(tt.keyToGet)
-		if ok != tt.expectedOk {
-			t.Fatalf("%s: cache hit = %v; want %v", tt.name, ok, !ok)
-		} else if ok && val != 1234 {
-			t.Fatalf("%s expected get to return 1234 but got %v", tt.name, val)
-		}
-	}
-}
-
-func TestRemove(t *testing.T) {
-	lru := New(0)
-	lru.Add("myKey", 1234)
-	if val, ok := lru.Get("myKey"); !ok {
-		t.Fatal("TestRemove returned no match")
-	} else if val != 1234 {
-		t.Fatalf("TestRemove failed.  Expected %d, got %v", 1234, val)
-	}
-
-	lru.Remove("myKey")
-	if _, ok := lru.Get("myKey"); ok {
-		t.Fatal("TestRemove returned a removed entry")
-	}
-}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/batch.go

@@ -8,65 +8,84 @@ package leveldb
 
 import (
 	"encoding/binary"
-	"errors"
+	"fmt"
 
+	"github.com/syndtr/goleveldb/leveldb/errors"
 	"github.com/syndtr/goleveldb/leveldb/memdb"
 )
 
-var (
-	errBatchTooShort  = errors.New("leveldb: batch is too short")
-	errBatchBadRecord = errors.New("leveldb: bad record in batch")
+type ErrBatchCorrupted struct {
+	Reason string
+}
+
+func (e *ErrBatchCorrupted) Error() string {
+	return fmt.Sprintf("leveldb: batch corrupted: %s", e.Reason)
+}
+
+func newErrBatchCorrupted(reason string) error {
+	return errors.NewErrCorrupted(nil, &ErrBatchCorrupted{reason})
+}
+
+const (
+	batchHdrLen  = 8 + 4
+	batchGrowRec = 3000
 )
 
-const kBatchHdrLen = 8 + 4
-
-type batchReplay interface {
-	put(key, value []byte, seq uint64)
-	delete(key []byte, seq uint64)
+type BatchReplay interface {
+	Put(key, value []byte)
+	Delete(key []byte)
 }
 
 // Batch is a write batch.
 type Batch struct {
-	buf        []byte
+	data       []byte
 	rLen, bLen int
 	seq        uint64
 	sync       bool
 }
 
 func (b *Batch) grow(n int) {
-	off := len(b.buf)
+	off := len(b.data)
 	if off == 0 {
-		// include headers
-		off = kBatchHdrLen
-		n += off
+		off = batchHdrLen
+		if b.data != nil {
+			b.data = b.data[:off]
+		}
 	}
-	if cap(b.buf)-off >= n {
-		return
+	if cap(b.data)-off < n {
+		if b.data == nil {
+			b.data = make([]byte, off, off+n)
+		} else {
+			odata := b.data
+			div := 1
+			if b.rLen > batchGrowRec {
+				div = b.rLen / batchGrowRec
+			}
+			b.data = make([]byte, off, off+n+(off-batchHdrLen)/div)
+			copy(b.data, odata)
+		}
 	}
-	buf := make([]byte, 2*cap(b.buf)+n)
-	copy(buf, b.buf)
-	b.buf = buf[:off]
 }
 
-func (b *Batch) appendRec(t vType, key, value []byte) {
+func (b *Batch) appendRec(kt kType, key, value []byte) {
 	n := 1 + binary.MaxVarintLen32 + len(key)
-	if t == tVal {
+	if kt == ktVal {
 		n += binary.MaxVarintLen32 + len(value)
 	}
 	b.grow(n)
-	off := len(b.buf)
-	buf := b.buf[:off+n]
-	buf[off] = byte(t)
+	off := len(b.data)
+	data := b.data[:off+n]
+	data[off] = byte(kt)
 	off += 1
-	off += binary.PutUvarint(buf[off:], uint64(len(key)))
-	copy(buf[off:], key)
+	off += binary.PutUvarint(data[off:], uint64(len(key)))
+	copy(data[off:], key)
 	off += len(key)
-	if t == tVal {
-		off += binary.PutUvarint(buf[off:], uint64(len(value)))
-		copy(buf[off:], value)
+	if kt == ktVal {
+		off += binary.PutUvarint(data[off:], uint64(len(value)))
+		copy(data[off:], value)
 		off += len(value)
 	}
-	b.buf = buf[:off]
+	b.data = data[:off]
 	b.rLen++
 	//  Include 8-byte ikey header
 	b.bLen += len(key) + len(value) + 8
@@ -75,18 +94,51 @@ func (b *Batch) appendRec(t vType, key, value []byte) {
 // Put appends 'put operation' of the given key/value pair to the batch.
 // It is safe to modify the contents of the argument after Put returns.
 func (b *Batch) Put(key, value []byte) {
-	b.appendRec(tVal, key, value)
+	b.appendRec(ktVal, key, value)
 }
 
 // Delete appends 'delete operation' of the given key to the batch.
 // It is safe to modify the contents of the argument after Delete returns.
 func (b *Batch) Delete(key []byte) {
-	b.appendRec(tDel, key, nil)
+	b.appendRec(ktDel, key, nil)
+}
+
+// Dump dumps batch contents. The returned slice can be loaded into the
+// batch using Load method.
+// The returned slice is not its own copy, so the contents should not be
+// modified.
+func (b *Batch) Dump() []byte {
+	return b.encode()
+}
+
+// Load loads given slice into the batch. Previous contents of the batch
+// will be discarded.
+// The given slice will not be copied and will be used as batch buffer, so
+// it is not safe to modify the contents of the slice.
+func (b *Batch) Load(data []byte) error {
+	return b.decode(0, data)
+}
+
+// Replay replays batch contents.
+func (b *Batch) Replay(r BatchReplay) error {
+	return b.decodeRec(func(i int, kt kType, key, value []byte) {
+		switch kt {
+		case ktVal:
+			r.Put(key, value)
+		case ktDel:
+			r.Delete(key)
+		}
+	})
+}
+
+// Len returns number of records in the batch.
+func (b *Batch) Len() int {
+	return b.rLen
 }
 
 // Reset resets the batch.
 func (b *Batch) Reset() {
-	b.buf = nil
+	b.data = b.data[:0]
 	b.seq = 0
 	b.rLen = 0
 	b.bLen = 0
@@ -97,24 +149,10 @@ func (b *Batch) init(sync bool) {
 	b.sync = sync
 }
 
-func (b *Batch) put(key, value []byte, seq uint64) {
-	if b.rLen == 0 {
-		b.seq = seq
-	}
-	b.Put(key, value)
-}
-
-func (b *Batch) delete(key []byte, seq uint64) {
-	if b.rLen == 0 {
-		b.seq = seq
-	}
-	b.Delete(key)
-}
-
 func (b *Batch) append(p *Batch) {
 	if p.rLen > 0 {
-		b.grow(len(p.buf) - kBatchHdrLen)
-		b.buf = append(b.buf, p.buf[kBatchHdrLen:]...)
+		b.grow(len(p.data) - batchHdrLen)
+		b.data = append(b.data, p.data[batchHdrLen:]...)
 		b.rLen += p.rLen
 	}
 	if p.sync {
@@ -122,95 +160,93 @@ func (b *Batch) append(p *Batch) {
 	}
 }
 
-func (b *Batch) len() int {
-	return b.rLen
-}
-
+// size returns sums of key/value pair length plus 8-bytes ikey.
 func (b *Batch) size() int {
 	return b.bLen
 }
 
 func (b *Batch) encode() []byte {
 	b.grow(0)
-	binary.LittleEndian.PutUint64(b.buf, b.seq)
-	binary.LittleEndian.PutUint32(b.buf[8:], uint32(b.rLen))
+	binary.LittleEndian.PutUint64(b.data, b.seq)
+	binary.LittleEndian.PutUint32(b.data[8:], uint32(b.rLen))
 
-	return b.buf
+	return b.data
 }
 
-func (b *Batch) decode(buf []byte) error {
-	if len(buf) < kBatchHdrLen {
-		return errBatchTooShort
+func (b *Batch) decode(prevSeq uint64, data []byte) error {
+	if len(data) < batchHdrLen {
+		return newErrBatchCorrupted("too short")
 	}
 
-	b.seq = binary.LittleEndian.Uint64(buf)
-	b.rLen = int(binary.LittleEndian.Uint32(buf[8:]))
+	b.seq = binary.LittleEndian.Uint64(data)
+	if b.seq < prevSeq {
+		return newErrBatchCorrupted("invalid sequence number")
+	}
+	b.rLen = int(binary.LittleEndian.Uint32(data[8:]))
+	if b.rLen < 0 {
+		return newErrBatchCorrupted("invalid records length")
+	}
 	// No need to be precise at this point, it won't be used anyway
-	b.bLen = len(buf) - kBatchHdrLen
-	b.buf = buf
+	b.bLen = len(data) - batchHdrLen
+	b.data = data
 
 	return nil
 }
 
-func (b *Batch) decodeRec(f func(i int, t vType, key, value []byte)) error {
-	off := kBatchHdrLen
+func (b *Batch) decodeRec(f func(i int, kt kType, key, value []byte)) (err error) {
+	off := batchHdrLen
 	for i := 0; i < b.rLen; i++ {
-		if off >= len(b.buf) {
-			return errors.New("leveldb: invalid batch record length")
+		if off >= len(b.data) {
+			return newErrBatchCorrupted("invalid records length")
 		}
 
-		t := vType(b.buf[off])
-		if t > tVal {
-			return errors.New("leveldb: invalid batch record type in batch")
+		kt := kType(b.data[off])
+		if kt > ktVal {
+			return newErrBatchCorrupted("bad record: invalid type")
 		}
 		off += 1
 
-		x, n := binary.Uvarint(b.buf[off:])
+		x, n := binary.Uvarint(b.data[off:])
 		off += n
-		if n <= 0 || off+int(x) > len(b.buf) {
-			return errBatchBadRecord
+		if n <= 0 || off+int(x) > len(b.data) {
+			return newErrBatchCorrupted("bad record: invalid key length")
 		}
-		key := b.buf[off : off+int(x)]
+		key := b.data[off : off+int(x)]
 		off += int(x)
-
 		var value []byte
-		if t == tVal {
-			x, n := binary.Uvarint(b.buf[off:])
+		if kt == ktVal {
+			x, n := binary.Uvarint(b.data[off:])
 			off += n
-			if n <= 0 || off+int(x) > len(b.buf) {
-				return errBatchBadRecord
+			if n <= 0 || off+int(x) > len(b.data) {
+				return newErrBatchCorrupted("bad record: invalid value length")
 			}
-			value = b.buf[off : off+int(x)]
+			value = b.data[off : off+int(x)]
 			off += int(x)
 		}
 
-		f(i, t, key, value)
+		f(i, kt, key, value)
 	}
 
 	return nil
 }
 
-func (b *Batch) replay(to batchReplay) error {
-	return b.decodeRec(func(i int, t vType, key, value []byte) {
-		switch t {
-		case tVal:
-			to.put(key, value, b.seq+uint64(i))
-		case tDel:
-			to.delete(key, b.seq+uint64(i))
-		}
-	})
-}
-
 func (b *Batch) memReplay(to *memdb.DB) error {
-	return b.decodeRec(func(i int, t vType, key, value []byte) {
-		ikey := newIKey(key, b.seq+uint64(i), t)
+	return b.decodeRec(func(i int, kt kType, key, value []byte) {
+		ikey := newIkey(key, b.seq+uint64(i), kt)
 		to.Put(ikey, value)
 	})
 }
 
+func (b *Batch) memDecodeAndReplay(prevSeq uint64, data []byte, to *memdb.DB) error {
+	if err := b.decode(prevSeq, data); err != nil {
+		return err
+	}
+	return b.memReplay(to)
+}
+
 func (b *Batch) revertMemReplay(to *memdb.DB) error {
-	return b.decodeRec(func(i int, t vType, key, value []byte) {
-		ikey := newIKey(key, b.seq+uint64(i), t)
+	return b.decodeRec(func(i int, kt kType, key, value []byte) {
+		ikey := newIkey(key, b.seq+uint64(i), kt)
 		to.Delete(ikey)
 	})
 }

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/batch_test.go

@@ -15,7 +15,7 @@ import (
 )
 
 type tbRec struct {
-	t          vType
+	kt         kType
 	key, value []byte
 }
 
@@ -23,39 +23,39 @@ type testBatch struct {
 	rec []*tbRec
 }
 
-func (p *testBatch) put(key, value []byte, seq uint64) {
-	p.rec = append(p.rec, &tbRec{tVal, key, value})
+func (p *testBatch) Put(key, value []byte) {
+	p.rec = append(p.rec, &tbRec{ktVal, key, value})
 }
 
-func (p *testBatch) delete(key []byte, seq uint64) {
-	p.rec = append(p.rec, &tbRec{tDel, key, nil})
+func (p *testBatch) Delete(key []byte) {
+	p.rec = append(p.rec, &tbRec{ktDel, key, nil})
 }
 
 func compareBatch(t *testing.T, b1, b2 *Batch) {
 	if b1.seq != b2.seq {
 		t.Errorf("invalid seq number want %d, got %d", b1.seq, b2.seq)
 	}
-	if b1.len() != b2.len() {
-		t.Fatalf("invalid record length want %d, got %d", b1.len(), b2.len())
+	if b1.Len() != b2.Len() {
+		t.Fatalf("invalid record length want %d, got %d", b1.Len(), b2.Len())
 	}
 	p1, p2 := new(testBatch), new(testBatch)
-	err := b1.replay(p1)
+	err := b1.Replay(p1)
 	if err != nil {
 		t.Fatal("error when replaying batch 1: ", err)
 	}
-	err = b2.replay(p2)
+	err = b2.Replay(p2)
 	if err != nil {
 		t.Fatal("error when replaying batch 2: ", err)
 	}
 	for i := range p1.rec {
 		r1, r2 := p1.rec[i], p2.rec[i]
-		if r1.t != r2.t {
-			t.Errorf("invalid type on record '%d' want %d, got %d", i, r1.t, r2.t)
+		if r1.kt != r2.kt {
+			t.Errorf("invalid type on record '%d' want %d, got %d", i, r1.kt, r2.kt)
 		}
 		if !bytes.Equal(r1.key, r2.key) {
 			t.Errorf("invalid key on record '%d' want %s, got %s", i, string(r1.key), string(r2.key))
 		}
-		if r1.t == tVal {
+		if r1.kt == ktVal {
 			if !bytes.Equal(r1.value, r2.value) {
 				t.Errorf("invalid value on record '%d' want %s, got %s", i, string(r1.value), string(r2.value))
 			}
@@ -75,7 +75,7 @@ func TestBatch_EncodeDecode(t *testing.T) {
 	b1.Delete([]byte("k"))
 	buf := b1.encode()
 	b2 := new(Batch)
-	err := b2.decode(buf)
+	err := b2.decode(0, buf)
 	if err != nil {
 		t.Error("error when decoding batch: ", err)
 	}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/bench_test.go

@@ -170,7 +170,7 @@ func (p *dbBench) writes(perBatch int) {
 	b.SetBytes(116)
 }
 
-func (p *dbBench) drop() {
+func (p *dbBench) gc() {
 	p.keys, p.values = nil, nil
 	runtime.GC()
 }
@@ -249,6 +249,9 @@ func (p *dbBench) newIter() iterator.Iterator {
 }
 
 func (p *dbBench) close() {
+	if bp, err := p.db.GetProperty("leveldb.blockpool"); err == nil {
+		p.b.Log("Block pool stats: ", bp)
+	}
 	p.db.Close()
 	p.stor.Close()
 	os.RemoveAll(benchDB)
@@ -331,7 +334,7 @@ func BenchmarkDBRead(b *testing.B) {
 	p := openDBBench(b, false)
 	p.populate(b.N)
 	p.fill()
-	p.drop()
+	p.gc()
 
 	iter := p.newIter()
 	b.ResetTimer()
@@ -362,7 +365,7 @@ func BenchmarkDBReadUncompressed(b *testing.B) {
 	p := openDBBench(b, true)
 	p.populate(b.N)
 	p.fill()
-	p.drop()
+	p.gc()
 
 	iter := p.newIter()
 	b.ResetTimer()
@@ -379,7 +382,7 @@ func BenchmarkDBReadTable(b *testing.B) {
 	p.populate(b.N)
 	p.fill()
 	p.reopen()
-	p.drop()
+	p.gc()
 
 	iter := p.newIter()
 	b.ResetTimer()
@@ -395,7 +398,7 @@ func BenchmarkDBReadReverse(b *testing.B) {
 	p := openDBBench(b, false)
 	p.populate(b.N)
 	p.fill()
-	p.drop()
+	p.gc()
 
 	iter := p.newIter()
 	b.ResetTimer()
@@ -413,7 +416,7 @@ func BenchmarkDBReadReverseTable(b *testing.B) {
 	p.populate(b.N)
 	p.fill()
 	p.reopen()
-	p.drop()
+	p.gc()
 
 	iter := p.newIter()
 	b.ResetTimer()

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/cache/cache.go

@@ -11,115 +11,149 @@ import (
 	"sync/atomic"
 )
 
-// SetFunc used by Namespace.Get method to create a cache object. SetFunc
-// may return ok false, in that case the cache object will not be created.
-type SetFunc func() (ok bool, value interface{}, charge int, fin SetFin)
+// SetFunc is the function that will be called by Namespace.Get to create
+// a cache object, if charge is less than one than the cache object will
+// not be registered to cache tree, if value is nil then the cache object
+// will not be created.
+type SetFunc func() (charge int, value interface{})
 
-// SetFin will be called when corresponding cache object are released.
-type SetFin func()
+// DelFin is the function that will be called as the result of a delete operation.
+// Exist == true is indication that the object is exist, and pending == true is
+// indication of deletion already happen but haven't done yet (wait for all handles
+// to be released). And exist == false means the object doesn't exist.
+type DelFin func(exist, pending bool)
 
-// DelFin will be called when corresponding cache object are released.
-// DelFin will be called after SetFin. The exist is true if the corresponding
-// cache object is actually exist in the cache tree.
-type DelFin func(exist bool)
+// PurgeFin is the function that will be called as the result of a purge operation.
+type PurgeFin func(ns, key uint64)
 
-// PurgeFin will be called when corresponding cache object are released.
-// PurgeFin will be called after SetFin. If PurgeFin present DelFin will
-// not be executed but passed to the PurgeFin, it is up to the caller
-// to call it or not.
-type PurgeFin func(ns, key uint64, delfin DelFin)
-
-// Cache is a cache tree.
+// Cache is a cache tree. A cache instance must be goroutine-safe.
 type Cache interface {
-	// SetCapacity sets cache capacity.
+	// SetCapacity sets cache tree capacity.
 	SetCapacity(capacity int)
 
-	// GetNamespace gets or creates a cache namespace for the given id.
+	// Capacity returns cache tree capacity.
+	Capacity() int
+
+	// Used returns used cache tree capacity.
+	Used() int
+
+	// Size returns entire alive cache objects size.
+	Size() int
+
+	// NumObjects returns number of alive objects.
+	NumObjects() int
+
+	// GetNamespace gets cache namespace with the given id.
+	// GetNamespace is never return nil.
 	GetNamespace(id uint64) Namespace
 
-	// Purge purges all cache namespaces, read Namespace.Purge method documentation.
+	// PurgeNamespace purges cache namespace with the given id from this cache tree.
+	// Also read Namespace.Purge.
+	PurgeNamespace(id uint64, fin PurgeFin)
+
+	// ZapNamespace detaches cache namespace with the given id from this cache tree.
+	// Also read Namespace.Zap.
+	ZapNamespace(id uint64)
+
+	// Purge purges all cache namespace from this cache tree.
+	// This is behave the same as calling Namespace.Purge method on all cache namespace.
 	Purge(fin PurgeFin)
 
-	// Zap zaps all cache namespaces, read Namespace.Zap method documentation.
-	Zap(closed bool)
+	// Zap detaches all cache namespace from this cache tree.
+	// This is behave the same as calling Namespace.Zap method on all cache namespace.
+	Zap()
 }
 
-// Namespace is a cache namespace.
+// Namespace is a cache namespace. A namespace instance must be goroutine-safe.
 type Namespace interface {
-	// Get gets cache object for the given key. The given SetFunc (if not nil) will
-	// be called if the given key does not exist.
-	// If the given key does not exist, SetFunc is nil or SetFunc return ok false, Get
-	// will return ok false.
-	Get(key uint64, setf SetFunc) (obj Object, ok bool)
-
-	// Get deletes cache object for the given key. If exist the cache object will
-	// be deleted later when all of its handles have been released (i.e. no one use
-	// it anymore) and the given DelFin (if not nil) will finally be  executed. If
-	// such cache object does not exist the given DelFin will be executed anyway.
+	// Get gets cache object with the given key.
+	// If cache object is not found and setf is not nil, Get will atomically creates
+	// the cache object by calling setf. Otherwise Get will returns nil.
 	//
-	// Delete returns true if such cache object exist.
+	// The returned cache handle should be released after use by calling Release
+	// method.
+	Get(key uint64, setf SetFunc) Handle
+
+	// Delete removes cache object with the given key from cache tree.
+	// A deleted cache object will be released as soon as all of its handles have
+	// been released.
+	// Delete only happen once, subsequent delete will consider cache object doesn't
+	// exist, even if the cache object ins't released yet.
+	//
+	// If not nil, fin will be called if the cache object doesn't exist or when
+	// finally be released.
+	//
+	// Delete returns true if such cache object exist and never been deleted.
 	Delete(key uint64, fin DelFin) bool
 
-	// Purge deletes all cache objects, read Delete method documentation.
+	// Purge removes all cache objects within this namespace from cache tree.
+	// This is the same as doing delete on all cache objects.
+	//
+	// If not nil, fin will be called on all cache objects when its finally be
+	// released.
 	Purge(fin PurgeFin)
 
-	// Zap detaches the namespace from the cache tree and delete all its cache
-	// objects. The cache objects deletion and finalizers execution are happen
-	// immediately, even if its existing handles haven't yet been released.
-	// A zapped namespace can't never be filled again.
-	// If closed is false then the Get function will always call the given SetFunc
-	// if it is not nil, but resultant of the SetFunc will not be cached.
-	Zap(closed bool)
+	// Zap detaches namespace from cache tree and release all its cache objects.
+	// A zapped namespace can never be filled again.
+	// Calling Get on zapped namespace will always return nil.
+	Zap()
 }
 
-// Object is a cache object.
-type Object interface {
-	// Release releases the cache object. Other methods should not be called
-	// after the cache object has been released.
+// Handle is a cache handle.
+type Handle interface {
+	// Release releases this cache handle. This method can be safely called mutiple
+	// times.
 	Release()
 
-	// Value returns value of the cache object.
+	// Value returns value of this cache handle.
+	// Value will returns nil after this cache handle have be released.
 	Value() interface{}
 }
 
+const (
+	DelNotExist = iota
+	DelExist
+	DelPendig
+)
+
 // Namespace state.
 type nsState int
 
 const (
 	nsEffective nsState = iota
 	nsZapped
-	nsClosed
 )
 
 // Node state.
 type nodeState int
 
 const (
-	nodeEffective nodeState = iota
+	nodeZero nodeState = iota
+	nodeEffective
 	nodeEvicted
-	nodeRemoved
+	nodeDeleted
 )
 
-// Fake object.
-type fakeObject struct {
+// Fake handle.
+type fakeHandle struct {
 	value interface{}
 	fin   func()
 	once  uint32
 }
 
-func (o *fakeObject) Value() interface{} {
-	if atomic.LoadUint32(&o.once) == 0 {
-		return o.value
+func (h *fakeHandle) Value() interface{} {
+	if atomic.LoadUint32(&h.once) == 0 {
+		return h.value
 	}
 	return nil
 }
 
-func (o *fakeObject) Release() {
-	if !atomic.CompareAndSwapUint32(&o.once, 0, 1) {
+func (h *fakeHandle) Release() {
+	if !atomic.CompareAndSwapUint32(&h.once, 0, 1) {
 		return
 	}
-	if o.fin != nil {
-		o.fin()
-		o.fin = nil
+	if h.fin != nil {
+		h.fin()
+		h.fin = nil
 	}
 }

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/cache/cache_test.go

@@ -8,14 +8,32 @@ package cache
 
 import (
 	"math/rand"
+	"runtime"
+	"sync"
+	"sync/atomic"
 	"testing"
+	"time"
 )
 
-func set(ns Namespace, key uint64, value interface{}, charge int, fin func()) Object {
-	obj, _ := ns.Get(key, func() (bool, interface{}, int, SetFin) {
-		return true, value, charge, fin
+type releaserFunc struct {
+	fn    func()
+	value interface{}
+}
+
+func (r releaserFunc) Release() {
+	if r.fn != nil {
+		r.fn()
+	}
+}
+
+func set(ns Namespace, key uint64, value interface{}, charge int, relf func()) Handle {
+	return ns.Get(key, func() (int, interface{}) {
+		if relf != nil {
+			return charge, releaserFunc{relf, value}
+		} else {
+			return charge, value
+		}
 	})
-	return obj
 }
 
 func TestCache_HitMiss(t *testing.T) {
@@ -43,29 +61,31 @@ func TestCache_HitMiss(t *testing.T) {
 			setfin++
 		}).Release()
 		for j, y := range cases {
-			r, ok := ns.Get(y.key, nil)
+			h := ns.Get(y.key, nil)
 			if j <= i {
 				// should hit
-				if !ok {
+				if h == nil {
 					t.Errorf("case '%d' iteration '%d' is miss", i, j)
-				} else if r.Value().(string) != y.value {
-					t.Errorf("case '%d' iteration '%d' has invalid value got '%s', want '%s'", i, j, r.Value().(string), y.value)
+				} else {
+					if x := h.Value().(releaserFunc).value.(string); x != y.value {
+						t.Errorf("case '%d' iteration '%d' has invalid value got '%s', want '%s'", i, j, x, y.value)
+					}
 				}
 			} else {
 				// should miss
-				if ok {
-					t.Errorf("case '%d' iteration '%d' is hit , value '%s'", i, j, r.Value().(string))
+				if h != nil {
+					t.Errorf("case '%d' iteration '%d' is hit , value '%s'", i, j, h.Value().(releaserFunc).value.(string))
 				}
 			}
-			if ok {
-				r.Release()
+			if h != nil {
+				h.Release()
 			}
 		}
 	}
 
 	for i, x := range cases {
 		finalizerOk := false
-		ns.Delete(x.key, func(exist bool) {
+		ns.Delete(x.key, func(exist, pending bool) {
 			finalizerOk = true
 		})
 
@@ -74,22 +94,24 @@ func TestCache_HitMiss(t *testing.T) {
 		}
 
 		for j, y := range cases {
-			r, ok := ns.Get(y.key, nil)
+			h := ns.Get(y.key, nil)
 			if j > i {
 				// should hit
-				if !ok {
+				if h == nil {
 					t.Errorf("case '%d' iteration '%d' is miss", i, j)
-				} else if r.Value().(string) != y.value {
-					t.Errorf("case '%d' iteration '%d' has invalid value got '%s', want '%s'", i, j, r.Value().(string), y.value)
+				} else {
+					if x := h.Value().(releaserFunc).value.(string); x != y.value {
+						t.Errorf("case '%d' iteration '%d' has invalid value got '%s', want '%s'", i, j, x, y.value)
+					}
 				}
 			} else {
 				// should miss
-				if ok {
-					t.Errorf("case '%d' iteration '%d' is hit, value '%s'", i, j, r.Value().(string))
+				if h != nil {
+					t.Errorf("case '%d' iteration '%d' is hit, value '%s'", i, j, h.Value().(releaserFunc).value.(string))
 				}
 			}
-			if ok {
-				r.Release()
+			if h != nil {
+				h.Release()
 			}
 		}
 	}
@@ -107,42 +129,42 @@ func TestLRUCache_Eviction(t *testing.T) {
 	set(ns, 3, 3, 1, nil).Release()
 	set(ns, 4, 4, 1, nil).Release()
 	set(ns, 5, 5, 1, nil).Release()
-	if r, ok := ns.Get(2, nil); ok { // 1,3,4,5,2
-		r.Release()
+	if h := ns.Get(2, nil); h != nil { // 1,3,4,5,2
+		h.Release()
 	}
 	set(ns, 9, 9, 10, nil).Release() // 5,2,9
 
-	for _, x := range []uint64{9, 2, 5, 1} {
-		r, ok := ns.Get(x, nil)
-		if !ok {
-			t.Errorf("miss for key '%d'", x)
+	for _, key := range []uint64{9, 2, 5, 1} {
+		h := ns.Get(key, nil)
+		if h == nil {
+			t.Errorf("miss for key '%d'", key)
 		} else {
-			if r.Value().(int) != int(x) {
-				t.Errorf("invalid value for key '%d' want '%d', got '%d'", x, x, r.Value().(int))
+			if x := h.Value().(int); x != int(key) {
+				t.Errorf("invalid value for key '%d' want '%d', got '%d'", key, key, x)
 			}
-			r.Release()
+			h.Release()
 		}
 	}
 	o1.Release()
-	for _, x := range []uint64{1, 2, 5} {
-		r, ok := ns.Get(x, nil)
-		if !ok {
-			t.Errorf("miss for key '%d'", x)
+	for _, key := range []uint64{1, 2, 5} {
+		h := ns.Get(key, nil)
+		if h == nil {
+			t.Errorf("miss for key '%d'", key)
 		} else {
-			if r.Value().(int) != int(x) {
-				t.Errorf("invalid value for key '%d' want '%d', got '%d'", x, x, r.Value().(int))
+			if x := h.Value().(int); x != int(key) {
+				t.Errorf("invalid value for key '%d' want '%d', got '%d'", key, key, x)
 			}
-			r.Release()
+			h.Release()
 		}
 	}
-	for _, x := range []uint64{3, 4, 9} {
-		r, ok := ns.Get(x, nil)
-		if ok {
-			t.Errorf("hit for key '%d'", x)
-			if r.Value().(int) != int(x) {
-				t.Errorf("invalid value for key '%d' want '%d', got '%d'", x, x, r.Value().(int))
+	for _, key := range []uint64{3, 4, 9} {
+		h := ns.Get(key, nil)
+		if h != nil {
+			t.Errorf("hit for key '%d'", key)
+			if x := h.Value().(int); x != int(key) {
+				t.Errorf("invalid value for key '%d' want '%d', got '%d'", key, key, x)
 			}
-			r.Release()
+			h.Release()
 		}
 	}
 }
@@ -153,16 +175,15 @@ func TestLRUCache_SetGet(t *testing.T) {
 	for i := 0; i < 200; i++ {
 		n := uint64(rand.Intn(99999) % 20)
 		set(ns, n, n, 1, nil).Release()
-		if p, ok := ns.Get(n, nil); ok {
-			if p.Value() == nil {
+		if h := ns.Get(n, nil); h != nil {
+			if h.Value() == nil {
 				t.Errorf("key '%d' contains nil value", n)
 			} else {
-				got := p.Value().(uint64)
-				if got != n {
-					t.Errorf("invalid value for key '%d' want '%d', got '%d'", n, n, got)
+				if x := h.Value().(uint64); x != n {
+					t.Errorf("invalid value for key '%d' want '%d', got '%d'", n, n, x)
 				}
 			}
-			p.Release()
+			h.Release()
 		} else {
 			t.Errorf("key '%d' doesn't exist", n)
 		}
@@ -176,31 +197,429 @@ func TestLRUCache_Purge(t *testing.T) {
 	o2 := set(ns1, 2, 2, 1, nil)
 	ns1.Purge(nil)
 	set(ns1, 3, 3, 1, nil).Release()
-	for _, x := range []uint64{1, 2, 3} {
-		r, ok := ns1.Get(x, nil)
-		if !ok {
-			t.Errorf("miss for key '%d'", x)
+	for _, key := range []uint64{1, 2, 3} {
+		h := ns1.Get(key, nil)
+		if h == nil {
+			t.Errorf("miss for key '%d'", key)
 		} else {
-			if r.Value().(int) != int(x) {
-				t.Errorf("invalid value for key '%d' want '%d', got '%d'", x, x, r.Value().(int))
+			if x := h.Value().(int); x != int(key) {
+				t.Errorf("invalid value for key '%d' want '%d', got '%d'", key, key, x)
 			}
-			r.Release()
+			h.Release()
 		}
 	}
 	o1.Release()
 	o2.Release()
-	for _, x := range []uint64{1, 2} {
-		r, ok := ns1.Get(x, nil)
-		if ok {
-			t.Errorf("hit for key '%d'", x)
-			if r.Value().(int) != int(x) {
-				t.Errorf("invalid value for key '%d' want '%d', got '%d'", x, x, r.Value().(int))
+	for _, key := range []uint64{1, 2} {
+		h := ns1.Get(key, nil)
+		if h != nil {
+			t.Errorf("hit for key '%d'", key)
+			if x := h.Value().(int); x != int(key) {
+				t.Errorf("invalid value for key '%d' want '%d', got '%d'", key, key, x)
 			}
-			r.Release()
+			h.Release()
 		}
 	}
 }
 
+type testingCacheObjectCounter struct {
+	created  uint
+	released uint
+}
+
+func (c *testingCacheObjectCounter) createOne() {
+	c.created++
+}
+
+func (c *testingCacheObjectCounter) releaseOne() {
+	c.released++
+}
+
+type testingCacheObject struct {
+	t   *testing.T
+	cnt *testingCacheObjectCounter
+
+	ns, key uint64
+
+	releaseCalled bool
+}
+
+func (x *testingCacheObject) Release() {
+	if !x.releaseCalled {
+		x.releaseCalled = true
+		x.cnt.releaseOne()
+	} else {
+		x.t.Errorf("duplicate setfin NS#%d KEY#%d", x.ns, x.key)
+	}
+}
+
+func TestLRUCache_ConcurrentSetGet(t *testing.T) {
+	runtime.GOMAXPROCS(runtime.NumCPU())
+
+	seed := time.Now().UnixNano()
+	t.Logf("seed=%d", seed)
+
+	const (
+		N = 2000000
+		M = 4000
+		C = 3
+	)
+
+	var set, get uint32
+
+	wg := &sync.WaitGroup{}
+	c := NewLRUCache(M / 4)
+	for ni := uint64(0); ni < C; ni++ {
+		r0 := rand.New(rand.NewSource(seed + int64(ni)))
+		r1 := rand.New(rand.NewSource(seed + int64(ni) + 1))
+		ns := c.GetNamespace(ni)
+
+		wg.Add(2)
+		go func(ns Namespace, r *rand.Rand) {
+			for i := 0; i < N; i++ {
+				x := uint64(r.Int63n(M))
+				o := ns.Get(x, func() (int, interface{}) {
+					atomic.AddUint32(&set, 1)
+					return 1, x
+				})
+				if v := o.Value().(uint64); v != x {
+					t.Errorf("#%d invalid value, got=%d", x, v)
+				}
+				o.Release()
+			}
+			wg.Done()
+		}(ns, r0)
+		go func(ns Namespace, r *rand.Rand) {
+			for i := 0; i < N; i++ {
+				x := uint64(r.Int63n(M))
+				o := ns.Get(x, nil)
+				if o != nil {
+					atomic.AddUint32(&get, 1)
+					if v := o.Value().(uint64); v != x {
+						t.Errorf("#%d invalid value, got=%d", x, v)
+					}
+					o.Release()
+				}
+			}
+			wg.Done()
+		}(ns, r1)
+	}
+
+	wg.Wait()
+
+	t.Logf("set=%d get=%d", set, get)
+}
+
+func TestLRUCache_Finalizer(t *testing.T) {
+	const (
+		capacity   = 100
+		goroutines = 100
+		iterations = 10000
+		keymax     = 8000
+	)
+
+	cnt := &testingCacheObjectCounter{}
+
+	c := NewLRUCache(capacity)
+
+	type instance struct {
+		seed       int64
+		rnd        *rand.Rand
+		nsid       uint64
+		ns         Namespace
+		effective  int
+		handles    []Handle
+		handlesMap map[uint64]int
+
+		delete          bool
+		purge           bool
+		zap             bool
+		wantDel         int
+		delfinCalled    int
+		delfinCalledAll int
+		delfinCalledEff int
+		purgefinCalled  int
+	}
+
+	instanceGet := func(p *instance, key uint64) {
+		h := p.ns.Get(key, func() (charge int, value interface{}) {
+			to := &testingCacheObject{
+				t: t, cnt: cnt,
+				ns:  p.nsid,
+				key: key,
+			}
+			p.effective++
+			cnt.createOne()
+			return 1, releaserFunc{func() {
+				to.Release()
+				p.effective--
+			}, to}
+		})
+		p.handles = append(p.handles, h)
+		p.handlesMap[key] = p.handlesMap[key] + 1
+	}
+	instanceRelease := func(p *instance, i int) {
+		h := p.handles[i]
+		key := h.Value().(releaserFunc).value.(*testingCacheObject).key
+		if n := p.handlesMap[key]; n == 0 {
+			t.Fatal("key ref == 0")
+		} else if n > 1 {
+			p.handlesMap[key] = n - 1
+		} else {
+			delete(p.handlesMap, key)
+		}
+		h.Release()
+		p.handles = append(p.handles[:i], p.handles[i+1:]...)
+		p.handles[len(p.handles) : len(p.handles)+1][0] = nil
+	}
+
+	seed := time.Now().UnixNano()
+	t.Logf("seed=%d", seed)
+
+	instances := make([]*instance, goroutines)
+	for i := range instances {
+		p := &instance{}
+		p.handlesMap = make(map[uint64]int)
+		p.seed = seed + int64(i)
+		p.rnd = rand.New(rand.NewSource(p.seed))
+		p.nsid = uint64(i)
+		p.ns = c.GetNamespace(p.nsid)
+		p.delete = i%6 == 0
+		p.purge = i%8 == 0
+		p.zap = i%12 == 0 || i%3 == 0
+		instances[i] = p
+	}
+
+	runr := rand.New(rand.NewSource(seed - 1))
+	run := func(rnd *rand.Rand, x []*instance, init func(p *instance) bool, fn func(p *instance, i int) bool) {
+		var (
+			rx []*instance
+			rn []int
+		)
+		if init == nil {
+			rx = append([]*instance{}, x...)
+			rn = make([]int, len(x))
+		} else {
+			for _, p := range x {
+				if init(p) {
+					rx = append(rx, p)
+					rn = append(rn, 0)
+				}
+			}
+		}
+		for len(rx) > 0 {
+			i := rand.Intn(len(rx))
+			if fn(rx[i], rn[i]) {
+				rn[i]++
+			} else {
+				rx = append(rx[:i], rx[i+1:]...)
+				rn = append(rn[:i], rn[i+1:]...)
+			}
+		}
+	}
+
+	// Get and release.
+	run(runr, instances, nil, func(p *instance, i int) bool {
+		if i < iterations {
+			if len(p.handles) == 0 || p.rnd.Int()%2 == 0 {
+				instanceGet(p, uint64(p.rnd.Intn(keymax)))
+			} else {
+				instanceRelease(p, p.rnd.Intn(len(p.handles)))
+			}
+			return true
+		} else {
+			return false
+		}
+	})
+
+	if used, cap := c.Used(), c.Capacity(); used > cap {
+		t.Errorf("Used > capacity, used=%d cap=%d", used, cap)
+	}
+
+	// Check effective objects.
+	for i, p := range instances {
+		if int(p.effective) < len(p.handlesMap) {
+			t.Errorf("#%d effective objects < acquired handle, eo=%d ah=%d", i, p.effective, len(p.handlesMap))
+		}
+	}
+
+	if want := int(cnt.created - cnt.released); c.Size() != want {
+		t.Errorf("Invalid cache size, want=%d got=%d", want, c.Size())
+	}
+
+	// First delete.
+	run(runr, instances, func(p *instance) bool {
+		p.wantDel = p.effective
+		return p.delete
+	}, func(p *instance, i int) bool {
+		key := uint64(i)
+		if key < keymax {
+			_, wantExist := p.handlesMap[key]
+			gotExist := p.ns.Delete(key, func(exist, pending bool) {
+				p.delfinCalledAll++
+				if exist {
+					p.delfinCalledEff++
+				}
+			})
+			if !gotExist && wantExist {
+				t.Errorf("delete on NS#%d KEY#%d not found", p.nsid, key)
+			}
+			return true
+		} else {
+			return false
+		}
+	})
+
+	// Second delete.
+	run(runr, instances, func(p *instance) bool {
+		p.delfinCalled = 0
+		return p.delete
+	}, func(p *instance, i int) bool {
+		key := uint64(i)
+		if key < keymax {
+			gotExist := p.ns.Delete(key, func(exist, pending bool) {
+				if exist && !pending {
+					t.Errorf("delete fin on NS#%d KEY#%d exist and not pending for deletion", p.nsid, key)
+				}
+				p.delfinCalled++
+			})
+			if gotExist {
+				t.Errorf("delete on NS#%d KEY#%d found", p.nsid, key)
+			}
+			return true
+		} else {
+			if p.delfinCalled != keymax {
+				t.Errorf("(2) NS#%d not all delete fin called, diff=%d", p.nsid, keymax-p.delfinCalled)
+			}
+			return false
+		}
+	})
+
+	// Purge.
+	run(runr, instances, func(p *instance) bool {
+		return p.purge
+	}, func(p *instance, i int) bool {
+		p.ns.Purge(func(ns, key uint64) {
+			p.purgefinCalled++
+		})
+		return false
+	})
+
+	if want := int(cnt.created - cnt.released); c.Size() != want {
+		t.Errorf("Invalid cache size, want=%d got=%d", want, c.Size())
+	}
+
+	// Release.
+	run(runr, instances, func(p *instance) bool {
+		return !p.zap
+	}, func(p *instance, i int) bool {
+		if len(p.handles) > 0 {
+			instanceRelease(p, len(p.handles)-1)
+			return true
+		} else {
+			return false
+		}
+	})
+
+	if want := int(cnt.created - cnt.released); c.Size() != want {
+		t.Errorf("Invalid cache size, want=%d got=%d", want, c.Size())
+	}
+
+	// Zap.
+	run(runr, instances, func(p *instance) bool {
+		return p.zap
+	}, func(p *instance, i int) bool {
+		p.ns.Zap()
+		p.handles = nil
+		p.handlesMap = nil
+		return false
+	})
+
+	if want := int(cnt.created - cnt.released); c.Size() != want {
+		t.Errorf("Invalid cache size, want=%d got=%d", want, c.Size())
+	}
+
+	if notrel, used := int(cnt.created-cnt.released), c.Used(); notrel != used {
+		t.Errorf("Invalid used value, want=%d got=%d", notrel, used)
+	}
+
+	c.Purge(nil)
+
+	for _, p := range instances {
+		if p.delete {
+			if p.delfinCalledAll != keymax {
+				t.Errorf("#%d not all delete fin called, purge=%v zap=%v diff=%d", p.nsid, p.purge, p.zap, keymax-p.delfinCalledAll)
+			}
+			if p.delfinCalledEff != p.wantDel {
+				t.Errorf("#%d not all effective delete fin called, diff=%d", p.nsid, p.wantDel-p.delfinCalledEff)
+			}
+			if p.purge && p.purgefinCalled > 0 {
+				t.Errorf("#%d some purge fin called, delete=%v zap=%v n=%d", p.nsid, p.delete, p.zap, p.purgefinCalled)
+			}
+		} else {
+			if p.purge {
+				if p.purgefinCalled != p.wantDel {
+					t.Errorf("#%d not all purge fin called, delete=%v zap=%v diff=%d", p.nsid, p.delete, p.zap, p.wantDel-p.purgefinCalled)
+				}
+			}
+		}
+	}
+
+	if cnt.created != cnt.released {
+		t.Errorf("Some cache object weren't released, created=%d released=%d", cnt.created, cnt.released)
+	}
+}
+
+func BenchmarkLRUCache_Set(b *testing.B) {
+	c := NewLRUCache(0)
+	ns := c.GetNamespace(0)
+	b.ResetTimer()
+	for i := uint64(0); i < uint64(b.N); i++ {
+		set(ns, i, "", 1, nil)
+	}
+}
+
+func BenchmarkLRUCache_Get(b *testing.B) {
+	c := NewLRUCache(0)
+	ns := c.GetNamespace(0)
+	b.ResetTimer()
+	for i := uint64(0); i < uint64(b.N); i++ {
+		set(ns, i, "", 1, nil)
+	}
+	b.ResetTimer()
+	for i := uint64(0); i < uint64(b.N); i++ {
+		ns.Get(i, nil)
+	}
+}
+
+func BenchmarkLRUCache_Get2(b *testing.B) {
+	c := NewLRUCache(0)
+	ns := c.GetNamespace(0)
+	b.ResetTimer()
+	for i := uint64(0); i < uint64(b.N); i++ {
+		set(ns, i, "", 1, nil)
+	}
+	b.ResetTimer()
+	for i := uint64(0); i < uint64(b.N); i++ {
+		ns.Get(i, func() (charge int, value interface{}) {
+			return 0, nil
+		})
+	}
+}
+
+func BenchmarkLRUCache_Release(b *testing.B) {
+	c := NewLRUCache(0)
+	ns := c.GetNamespace(0)
+	handles := make([]Handle, b.N)
+	for i := uint64(0); i < uint64(b.N); i++ {
+		handles[i] = set(ns, i, "", 1, nil)
+	}
+	b.ResetTimer()
+	for _, h := range handles {
+		h.Release()
+	}
+}
+
 func BenchmarkLRUCache_SetRelease(b *testing.B) {
 	capacity := b.N / 100
 	if capacity <= 0 {
@@ -210,7 +629,7 @@ func BenchmarkLRUCache_SetRelease(b *testing.B) {
 	ns := c.GetNamespace(0)
 	b.ResetTimer()
 	for i := uint64(0); i < uint64(b.N); i++ {
-		set(ns, i, nil, 1, nil).Release()
+		set(ns, i, "", 1, nil).Release()
 	}
 }
 
@@ -227,10 +646,10 @@ func BenchmarkLRUCache_SetReleaseTwice(b *testing.B) {
 	nb := b.N - na
 
 	for i := uint64(0); i < uint64(na); i++ {
-		set(ns, i, nil, 1, nil).Release()
+		set(ns, i, "", 1, nil).Release()
 	}
 
 	for i := uint64(0); i < uint64(nb); i++ {
-		set(ns, i, nil, 1, nil).Release()
+		set(ns, i, "", 1, nil).Release()
 	}
 }

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/cache/empty_cache.go

@@ -1,246 +0,0 @@
-// Copyright (c) 2013, Suryandaru Triandana <syndtr@gmail.com>
-// All rights reserved.
-//
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-
-package cache
-
-import (
-	"sync"
-	"sync/atomic"
-)
-
-type emptyCache struct {
-	sync.Mutex
-	table map[uint64]*emptyNS
-}
-
-// NewEmptyCache creates a new initialized empty cache.
-func NewEmptyCache() Cache {
-	return &emptyCache{
-		table: make(map[uint64]*emptyNS),
-	}
-}
-
-func (c *emptyCache) GetNamespace(id uint64) Namespace {
-	c.Lock()
-	defer c.Unlock()
-
-	if ns, ok := c.table[id]; ok {
-		return ns
-	}
-
-	ns := &emptyNS{
-		cache: c,
-		id:    id,
-		table: make(map[uint64]*emptyNode),
-	}
-	c.table[id] = ns
-	return ns
-}
-
-func (c *emptyCache) Purge(fin PurgeFin) {
-	c.Lock()
-	for _, ns := range c.table {
-		ns.purgeNB(fin)
-	}
-	c.Unlock()
-}
-
-func (c *emptyCache) Zap(closed bool) {
-	c.Lock()
-	for _, ns := range c.table {
-		ns.zapNB(closed)
-	}
-	c.table = make(map[uint64]*emptyNS)
-	c.Unlock()
-}
-
-func (*emptyCache) SetCapacity(capacity int) {}
-
-type emptyNS struct {
-	cache *emptyCache
-	id    uint64
-	table map[uint64]*emptyNode
-	state nsState
-}
-
-func (ns *emptyNS) Get(key uint64, setf SetFunc) (o Object, ok bool) {
-	ns.cache.Lock()
-
-	switch ns.state {
-	case nsZapped:
-		ns.cache.Unlock()
-		if setf == nil {
-			return
-		}
-
-		var value interface{}
-		var fin func()
-		ok, value, _, fin = setf()
-		if ok {
-			o = &fakeObject{
-				value: value,
-				fin:   fin,
-			}
-		}
-		return
-	case nsClosed:
-		ns.cache.Unlock()
-		return
-	}
-
-	n, ok := ns.table[key]
-	if ok {
-		n.ref++
-	} else {
-		if setf == nil {
-			ns.cache.Unlock()
-			return
-		}
-
-		var value interface{}
-		var fin func()
-		ok, value, _, fin = setf()
-		if !ok {
-			ns.cache.Unlock()
-			return
-		}
-
-		n = &emptyNode{
-			ns:     ns,
-			key:    key,
-			value:  value,
-			setfin: fin,
-			ref:    1,
-		}
-		ns.table[key] = n
-	}
-
-	ns.cache.Unlock()
-	o = &emptyObject{node: n}
-	return
-}
-
-func (ns *emptyNS) Delete(key uint64, fin DelFin) bool {
-	ns.cache.Lock()
-
-	if ns.state != nsEffective {
-		ns.cache.Unlock()
-		if fin != nil {
-			fin(false)
-		}
-		return false
-	}
-
-	n, ok := ns.table[key]
-	if !ok {
-		ns.cache.Unlock()
-		if fin != nil {
-			fin(false)
-		}
-		return false
-	}
-	n.delfin = fin
-	ns.cache.Unlock()
-	return true
-}
-
-func (ns *emptyNS) purgeNB(fin PurgeFin) {
-	if ns.state != nsEffective {
-		return
-	}
-	for _, n := range ns.table {
-		n.purgefin = fin
-	}
-}
-
-func (ns *emptyNS) Purge(fin PurgeFin) {
-	ns.cache.Lock()
-	ns.purgeNB(fin)
-	ns.cache.Unlock()
-}
-
-func (ns *emptyNS) zapNB(closed bool) {
-	if ns.state != nsEffective {
-		return
-	}
-	for _, n := range ns.table {
-		n.execFin()
-	}
-	if closed {
-		ns.state = nsClosed
-	} else {
-		ns.state = nsZapped
-	}
-	ns.table = nil
-}
-
-func (ns *emptyNS) Zap(closed bool) {
-	ns.cache.Lock()
-	ns.zapNB(closed)
-	delete(ns.cache.table, ns.id)
-	ns.cache.Unlock()
-}
-
-type emptyNode struct {
-	ns       *emptyNS
-	key      uint64
-	value    interface{}
-	ref      int
-	setfin   SetFin
-	delfin   DelFin
-	purgefin PurgeFin
-}
-
-func (n *emptyNode) execFin() {
-	if n.setfin != nil {
-		n.setfin()
-		n.setfin = nil
-	}
-	if n.purgefin != nil {
-		n.purgefin(n.ns.id, n.key, n.delfin)
-		n.delfin = nil
-		n.purgefin = nil
-	} else if n.delfin != nil {
-		n.delfin(true)
-		n.delfin = nil
-	}
-}
-
-func (n *emptyNode) evict() {
-	n.ns.cache.Lock()
-	n.ref--
-	if n.ref == 0 {
-		if n.ns.state == nsEffective {
-			// Remove elem.
-			delete(n.ns.table, n.key)
-			// Execute finalizer.
-			n.execFin()
-		}
-	} else if n.ref < 0 {
-		panic("leveldb/cache: emptyNode: negative node reference")
-	}
-	n.ns.cache.Unlock()
-}
-
-type emptyObject struct {
-	node *emptyNode
-	once uint32
-}
-
-func (o *emptyObject) Value() interface{} {
-	if atomic.LoadUint32(&o.once) == 0 {
-		return o.node.value
-	}
-	return nil
-}
-
-func (o *emptyObject) Release() {
-	if !atomic.CompareAndSwapUint32(&o.once, 0, 1) {
-		return
-	}
-	o.node.evict()
-	o.node = nil
-}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/cache/lru_cache.go

@@ -9,16 +9,24 @@ package cache
 import (
 	"sync"
 	"sync/atomic"
+
+	"github.com/syndtr/goleveldb/leveldb/util"
 )
 
+// The LLRB implementation were taken from https://github.com/petar/GoLLRB.
+// Which contains the following header:
+//
+// Copyright 2010 Petar Maymounkov. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
 // lruCache represent a LRU cache state.
 type lruCache struct {
-	sync.Mutex
-
-	recent   lruNode
-	table    map[uint64]*lruNs
-	capacity int
-	size     int
+	mu                sync.Mutex
+	recent            lruNode
+	table             map[uint64]*lruNs
+	capacity          int
+	used, size, alive int
 }
 
 // NewLRUCache creates a new initialized LRU cache with the given capacity.
@@ -32,245 +40,415 @@ func NewLRUCache(capacity int) Cache {
 	return c
 }
 
+func (c *lruCache) Capacity() int {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	return c.capacity
+}
+
+func (c *lruCache) Used() int {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	return c.used
+}
+
+func (c *lruCache) Size() int {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	return c.size
+}
+
+func (c *lruCache) NumObjects() int {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	return c.alive
+}
+
 // SetCapacity set cache capacity.
 func (c *lruCache) SetCapacity(capacity int) {
-	c.Lock()
+	c.mu.Lock()
 	c.capacity = capacity
 	c.evict()
-	c.Unlock()
+	c.mu.Unlock()
 }
 
 // GetNamespace return namespace object for given id.
 func (c *lruCache) GetNamespace(id uint64) Namespace {
-	c.Lock()
-	defer c.Unlock()
+	c.mu.Lock()
+	defer c.mu.Unlock()
 
-	if p, ok := c.table[id]; ok {
-		return p
+	if ns, ok := c.table[id]; ok {
+		return ns
 	}
 
-	p := &lruNs{
-		lru:   c,
-		id:    id,
-		table: make(map[uint64]*lruNode),
+	ns := &lruNs{lru: c, id: id}
+	c.table[id] = ns
+	return ns
+}
+
+func (c *lruCache) ZapNamespace(id uint64) {
+	c.mu.Lock()
+	if ns, exist := c.table[id]; exist {
+		ns.zapNB()
+		delete(c.table, id)
 	}
-	c.table[id] = p
-	return p
+	c.mu.Unlock()
+}
+
+func (c *lruCache) PurgeNamespace(id uint64, fin PurgeFin) {
+	c.mu.Lock()
+	if ns, exist := c.table[id]; exist {
+		ns.purgeNB(fin)
+	}
+	c.mu.Unlock()
 }
 
 // Purge purge entire cache.
 func (c *lruCache) Purge(fin PurgeFin) {
-	c.Lock()
+	c.mu.Lock()
 	for _, ns := range c.table {
 		ns.purgeNB(fin)
 	}
-	c.Unlock()
+	c.mu.Unlock()
 }
 
-func (c *lruCache) Zap(closed bool) {
-	c.Lock()
+func (c *lruCache) Zap() {
+	c.mu.Lock()
 	for _, ns := range c.table {
-		ns.zapNB(closed)
+		ns.zapNB()
 	}
 	c.table = make(map[uint64]*lruNs)
-	c.Unlock()
+	c.mu.Unlock()
 }
 
 func (c *lruCache) evict() {
 	top := &c.recent
-	for n := c.recent.rPrev; c.size > c.capacity && n != top; {
+	for n := c.recent.rPrev; c.used > c.capacity && n != top; {
+		if n.state != nodeEffective {
+			panic("evicting non effective node")
+		}
 		n.state = nodeEvicted
 		n.rRemove()
-		n.evictNB()
-		c.size -= n.charge
+		n.derefNB()
+		c.used -= n.charge
 		n = c.recent.rPrev
 	}
 }
 
 type lruNs struct {
-	lru   *lruCache
-	id    uint64
-	table map[uint64]*lruNode
-	state nsState
+	lru    *lruCache
+	id     uint64
+	rbRoot *lruNode
+	state  nsState
 }
 
-func (ns *lruNs) Get(key uint64, setf SetFunc) (o Object, ok bool) {
-	lru := ns.lru
-	lru.Lock()
-
-	switch ns.state {
-	case nsZapped:
-		lru.Unlock()
-		if setf == nil {
-			return
-		}
-
-		var value interface{}
-		var fin func()
-		ok, value, _, fin = setf()
-		if ok {
-			o = &fakeObject{
-				value: value,
-				fin:   fin,
-			}
-		}
-		return
-	case nsClosed:
-		lru.Unlock()
-		return
+func (ns *lruNs) rbGetOrCreateNode(h *lruNode, key uint64) (hn, n *lruNode) {
+	if h == nil {
+		n = &lruNode{ns: ns, key: key}
+		return n, n
 	}
 
-	n, ok := ns.table[key]
-	if ok {
-		switch n.state {
-		case nodeEvicted:
-			// Insert to recent list.
-			n.state = nodeEffective
-			n.ref++
-			lru.size += n.charge
-			lru.evict()
-			fallthrough
-		case nodeEffective:
-			// Bump to front
-			n.rRemove()
-			n.rInsert(&lru.recent)
+	if key < h.key {
+		hn, n = ns.rbGetOrCreateNode(h.rbLeft, key)
+		if hn != nil {
+			h.rbLeft = hn
+		} else {
+			return nil, n
+		}
+	} else if key > h.key {
+		hn, n = ns.rbGetOrCreateNode(h.rbRight, key)
+		if hn != nil {
+			h.rbRight = hn
+		} else {
+			return nil, n
 		}
-		n.ref++
 	} else {
-		if setf == nil {
-			lru.Unlock()
-			return
-		}
-
-		var value interface{}
-		var charge int
-		var fin func()
-		ok, value, charge, fin = setf()
-		if !ok {
-			lru.Unlock()
-			return
-		}
-
-		n = &lruNode{
-			ns:     ns,
-			key:    key,
-			value:  value,
-			charge: charge,
-			setfin: fin,
-			ref:    2,
-		}
-		ns.table[key] = n
-		n.rInsert(&lru.recent)
-
-		lru.size += charge
-		lru.evict()
+		return nil, h
 	}
 
-	lru.Unlock()
-	o = &lruObject{node: n}
-	return
+	if rbIsRed(h.rbRight) && !rbIsRed(h.rbLeft) {
+		h = rbRotLeft(h)
+	}
+	if rbIsRed(h.rbLeft) && rbIsRed(h.rbLeft.rbLeft) {
+		h = rbRotRight(h)
+	}
+	if rbIsRed(h.rbLeft) && rbIsRed(h.rbRight) {
+		rbFlip(h)
+	}
+	return h, n
+}
+
+func (ns *lruNs) getOrCreateNode(key uint64) *lruNode {
+	hn, n := ns.rbGetOrCreateNode(ns.rbRoot, key)
+	if hn != nil {
+		ns.rbRoot = hn
+		ns.rbRoot.rbBlack = true
+	}
+	return n
+}
+
+func (ns *lruNs) rbGetNode(key uint64) *lruNode {
+	h := ns.rbRoot
+	for h != nil {
+		switch {
+		case key < h.key:
+			h = h.rbLeft
+		case key > h.key:
+			h = h.rbRight
+		default:
+			return h
+		}
+	}
+	return nil
+}
+
+func (ns *lruNs) getNode(key uint64) *lruNode {
+	return ns.rbGetNode(key)
+}
+
+func (ns *lruNs) rbDeleteNode(h *lruNode, key uint64) *lruNode {
+	if h == nil {
+		return nil
+	}
+
+	if key < h.key {
+		if h.rbLeft == nil { // key not present. Nothing to delete
+			return h
+		}
+		if !rbIsRed(h.rbLeft) && !rbIsRed(h.rbLeft.rbLeft) {
+			h = rbMoveLeft(h)
+		}
+		h.rbLeft = ns.rbDeleteNode(h.rbLeft, key)
+	} else {
+		if rbIsRed(h.rbLeft) {
+			h = rbRotRight(h)
+		}
+		// If @key equals @h.key and no right children at @h
+		if h.key == key && h.rbRight == nil {
+			return nil
+		}
+		if h.rbRight != nil && !rbIsRed(h.rbRight) && !rbIsRed(h.rbRight.rbLeft) {
+			h = rbMoveRight(h)
+		}
+		// If @key equals @h.key, and (from above) 'h.Right != nil'
+		if h.key == key {
+			var x *lruNode
+			h.rbRight, x = rbDeleteMin(h.rbRight)
+			if x == nil {
+				panic("logic")
+			}
+			x.rbLeft, h.rbLeft = h.rbLeft, nil
+			x.rbRight, h.rbRight = h.rbRight, nil
+			x.rbBlack = h.rbBlack
+			h = x
+		} else { // Else, @key is bigger than @h.key
+			h.rbRight = ns.rbDeleteNode(h.rbRight, key)
+		}
+	}
+
+	return rbFixup(h)
+}
+
+func (ns *lruNs) deleteNode(key uint64) {
+	ns.rbRoot = ns.rbDeleteNode(ns.rbRoot, key)
+	if ns.rbRoot != nil {
+		ns.rbRoot.rbBlack = true
+	}
+}
+
+func (ns *lruNs) rbIterateNodes(h *lruNode, pivot uint64, iter func(n *lruNode) bool) bool {
+	if h == nil {
+		return true
+	}
+	if h.key >= pivot {
+		if !ns.rbIterateNodes(h.rbLeft, pivot, iter) {
+			return false
+		}
+		if !iter(h) {
+			return false
+		}
+	}
+	return ns.rbIterateNodes(h.rbRight, pivot, iter)
+}
+
+func (ns *lruNs) iterateNodes(iter func(n *lruNode) bool) {
+	ns.rbIterateNodes(ns.rbRoot, 0, iter)
+}
+
+func (ns *lruNs) Get(key uint64, setf SetFunc) Handle {
+	ns.lru.mu.Lock()
+	defer ns.lru.mu.Unlock()
+
+	if ns.state != nsEffective {
+		return nil
+	}
+
+	var n *lruNode
+	if setf == nil {
+		n = ns.getNode(key)
+		if n == nil {
+			return nil
+		}
+	} else {
+		n = ns.getOrCreateNode(key)
+	}
+	switch n.state {
+	case nodeZero:
+		charge, value := setf()
+		if value == nil {
+			ns.deleteNode(key)
+			return nil
+		}
+		if charge < 0 {
+			charge = 0
+		}
+
+		n.value = value
+		n.charge = charge
+		n.state = nodeEvicted
+
+		ns.lru.size += charge
+		ns.lru.alive++
+
+		fallthrough
+	case nodeEvicted:
+		if n.charge == 0 {
+			break
+		}
+
+		// Insert to recent list.
+		n.state = nodeEffective
+		n.ref++
+		ns.lru.used += n.charge
+		ns.lru.evict()
+
+		fallthrough
+	case nodeEffective:
+		// Bump to front.
+		n.rRemove()
+		n.rInsert(&ns.lru.recent)
+	case nodeDeleted:
+		// Do nothing.
+	default:
+		panic("invalid state")
+	}
+	n.ref++
+
+	return &lruHandle{node: n}
 }
 
 func (ns *lruNs) Delete(key uint64, fin DelFin) bool {
-	lru := ns.lru
-	lru.Lock()
+	ns.lru.mu.Lock()
+	defer ns.lru.mu.Unlock()
 
 	if ns.state != nsEffective {
-		lru.Unlock()
 		if fin != nil {
-			fin(false)
+			fin(false, false)
 		}
 		return false
 	}
 
-	n, ok := ns.table[key]
-	if !ok {
-		lru.Unlock()
+	n := ns.getNode(key)
+	if n == nil {
 		if fin != nil {
-			fin(false)
+			fin(false, false)
 		}
 		return false
+
 	}
 
-	n.delfin = fin
 	switch n.state {
-	case nodeRemoved:
-		lru.Unlock()
-		return false
 	case nodeEffective:
-		lru.size -= n.charge
+		ns.lru.used -= n.charge
+		n.state = nodeDeleted
+		n.delfin = fin
 		n.rRemove()
-		n.evictNB()
+		n.derefNB()
+	case nodeEvicted:
+		n.state = nodeDeleted
+		n.delfin = fin
+	case nodeDeleted:
+		if fin != nil {
+			fin(true, true)
+		}
+		return false
+	default:
+		panic("invalid state")
 	}
-	n.state = nodeRemoved
 
-	lru.Unlock()
 	return true
 }
 
 func (ns *lruNs) purgeNB(fin PurgeFin) {
-	lru := ns.lru
-	if ns.state != nsEffective {
-		return
-	}
-
-	for _, n := range ns.table {
-		n.purgefin = fin
-		if n.state == nodeEffective {
-			lru.size -= n.charge
-			n.rRemove()
-			n.evictNB()
+	if ns.state == nsEffective {
+		var nodes []*lruNode
+		ns.iterateNodes(func(n *lruNode) bool {
+			nodes = append(nodes, n)
+			return true
+		})
+		for _, n := range nodes {
+			switch n.state {
+			case nodeEffective:
+				ns.lru.used -= n.charge
+				n.state = nodeDeleted
+				n.purgefin = fin
+				n.rRemove()
+				n.derefNB()
+			case nodeEvicted:
+				n.state = nodeDeleted
+				n.purgefin = fin
+			case nodeDeleted:
+			default:
+				panic("invalid state")
+			}
 		}
-		n.state = nodeRemoved
 	}
 }
 
 func (ns *lruNs) Purge(fin PurgeFin) {
-	ns.lru.Lock()
+	ns.lru.mu.Lock()
 	ns.purgeNB(fin)
-	ns.lru.Unlock()
+	ns.lru.mu.Unlock()
 }
 
-func (ns *lruNs) zapNB(closed bool) {
-	lru := ns.lru
-	if ns.state != nsEffective {
-		return
-	}
-
-	if closed {
-		ns.state = nsClosed
-	} else {
+func (ns *lruNs) zapNB() {
+	if ns.state == nsEffective {
 		ns.state = nsZapped
+
+		ns.iterateNodes(func(n *lruNode) bool {
+			if n.state == nodeEffective {
+				ns.lru.used -= n.charge
+				n.rRemove()
+			}
+			ns.lru.size -= n.charge
+			n.state = nodeDeleted
+			n.fin()
+
+			return true
+		})
+		ns.rbRoot = nil
 	}
-	for _, n := range ns.table {
-		if n.state == nodeEffective {
-			lru.size -= n.charge
-			n.rRemove()
-		}
-		n.state = nodeRemoved
-		n.execFin()
-	}
-	ns.table = nil
 }
 
-func (ns *lruNs) Zap(closed bool) {
-	ns.lru.Lock()
-	ns.zapNB(closed)
+func (ns *lruNs) Zap() {
+	ns.lru.mu.Lock()
+	ns.zapNB()
 	delete(ns.lru.table, ns.id)
-	ns.lru.Unlock()
+	ns.lru.mu.Unlock()
 }
 
 type lruNode struct {
 	ns *lruNs
 
-	rNext, rPrev *lruNode
+	rNext, rPrev    *lruNode
+	rbLeft, rbRight *lruNode
+	rbBlack         bool
 
 	key      uint64
 	value    interface{}
 	charge   int
 	ref      int
 	state    nodeState
-	setfin   SetFin
 	delfin   DelFin
 	purgefin PurgeFin
 }
@@ -284,7 +462,6 @@ func (n *lruNode) rInsert(at *lruNode) {
 }
 
 func (n *lruNode) rRemove() bool {
-	// only remove if not already removed
 	if n.rPrev == nil {
 		return false
 	}
@@ -297,58 +474,149 @@ func (n *lruNode) rRemove() bool {
 	return true
 }
 
-func (n *lruNode) execFin() {
-	if n.setfin != nil {
-		n.setfin()
-		n.setfin = nil
+func (n *lruNode) fin() {
+	if r, ok := n.value.(util.Releaser); ok {
+		r.Release()
 	}
 	if n.purgefin != nil {
-		n.purgefin(n.ns.id, n.key, n.delfin)
-		n.delfin = nil
+		if n.delfin != nil {
+			panic("conflicting delete and purge fin")
+		}
+		n.purgefin(n.ns.id, n.key)
 		n.purgefin = nil
 	} else if n.delfin != nil {
-		n.delfin(true)
+		n.delfin(true, false)
 		n.delfin = nil
 	}
 }
 
-func (n *lruNode) evictNB() {
+func (n *lruNode) derefNB() {
 	n.ref--
 	if n.ref == 0 {
 		if n.ns.state == nsEffective {
-			// remove elem
-			delete(n.ns.table, n.key)
-			// execute finalizer
-			n.execFin()
+			// Remove elemement.
+			n.ns.deleteNode(n.key)
+			n.ns.lru.size -= n.charge
+			n.ns.lru.alive--
+			n.fin()
 		}
+		n.value = nil
 	} else if n.ref < 0 {
 		panic("leveldb/cache: lruCache: negative node reference")
 	}
 }
 
-func (n *lruNode) evict() {
-	n.ns.lru.Lock()
-	n.evictNB()
-	n.ns.lru.Unlock()
+func (n *lruNode) deref() {
+	n.ns.lru.mu.Lock()
+	n.derefNB()
+	n.ns.lru.mu.Unlock()
 }
 
-type lruObject struct {
+type lruHandle struct {
 	node *lruNode
 	once uint32
 }
 
-func (o *lruObject) Value() interface{} {
-	if atomic.LoadUint32(&o.once) == 0 {
-		return o.node.value
+func (h *lruHandle) Value() interface{} {
+	if atomic.LoadUint32(&h.once) == 0 {
+		return h.node.value
 	}
 	return nil
 }
 
-func (o *lruObject) Release() {
-	if !atomic.CompareAndSwapUint32(&o.once, 0, 1) {
+func (h *lruHandle) Release() {
+	if !atomic.CompareAndSwapUint32(&h.once, 0, 1) {
 		return
 	}
-
-	o.node.evict()
-	o.node = nil
+	h.node.deref()
+	h.node = nil
+}
+
+func rbIsRed(h *lruNode) bool {
+	if h == nil {
+		return false
+	}
+	return !h.rbBlack
+}
+
+func rbRotLeft(h *lruNode) *lruNode {
+	x := h.rbRight
+	if x.rbBlack {
+		panic("rotating a black link")
+	}
+	h.rbRight = x.rbLeft
+	x.rbLeft = h
+	x.rbBlack = h.rbBlack
+	h.rbBlack = false
+	return x
+}
+
+func rbRotRight(h *lruNode) *lruNode {
+	x := h.rbLeft
+	if x.rbBlack {
+		panic("rotating a black link")
+	}
+	h.rbLeft = x.rbRight
+	x.rbRight = h
+	x.rbBlack = h.rbBlack
+	h.rbBlack = false
+	return x
+}
+
+func rbFlip(h *lruNode) {
+	h.rbBlack = !h.rbBlack
+	h.rbLeft.rbBlack = !h.rbLeft.rbBlack
+	h.rbRight.rbBlack = !h.rbRight.rbBlack
+}
+
+func rbMoveLeft(h *lruNode) *lruNode {
+	rbFlip(h)
+	if rbIsRed(h.rbRight.rbLeft) {
+		h.rbRight = rbRotRight(h.rbRight)
+		h = rbRotLeft(h)
+		rbFlip(h)
+	}
+	return h
+}
+
+func rbMoveRight(h *lruNode) *lruNode {
+	rbFlip(h)
+	if rbIsRed(h.rbLeft.rbLeft) {
+		h = rbRotRight(h)
+		rbFlip(h)
+	}
+	return h
+}
+
+func rbFixup(h *lruNode) *lruNode {
+	if rbIsRed(h.rbRight) {
+		h = rbRotLeft(h)
+	}
+
+	if rbIsRed(h.rbLeft) && rbIsRed(h.rbLeft.rbLeft) {
+		h = rbRotRight(h)
+	}
+
+	if rbIsRed(h.rbLeft) && rbIsRed(h.rbRight) {
+		rbFlip(h)
+	}
+
+	return h
+}
+
+func rbDeleteMin(h *lruNode) (hn, n *lruNode) {
+	if h == nil {
+		return nil, nil
+	}
+	if h.rbLeft == nil {
+		return nil, h
+	}
+
+	if !rbIsRed(h.rbLeft) && !rbIsRed(h.rbLeft.rbLeft) {
+		h = rbMoveLeft(h)
+	}
+
+	h.rbLeft, n = rbDeleteMin(h.rbLeft)
+
+	return rbFixup(h), n
 }

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/config.go

@@ -1,40 +0,0 @@
-// Copyright (c) 2012, Suryandaru Triandana <syndtr@gmail.com>
-// All rights reserved.
-//
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-
-package leveldb
-
-const (
-	kNumLevels = 7
-
-	// Level-0 compaction is started when we hit this many files.
-	kL0_CompactionTrigger float64 = 4
-
-	// Soft limit on number of level-0 files.  We slow down writes at this point.
-	kL0_SlowdownWritesTrigger = 8
-
-	// Maximum number of level-0 files.  We stop writes at this point.
-	kL0_StopWritesTrigger = 12
-
-	// Maximum level to which a new compacted memdb is pushed if it
-	// does not create overlap.  We try to push to level 2 to avoid the
-	// relatively expensive level 0=>1 compactions and to avoid some
-	// expensive manifest file operations.  We do not push all the way to
-	// the largest level since that can generate a lot of wasted disk
-	// space if the same key space is being repeatedly overwritten.
-	kMaxMemCompactLevel = 2
-
-	// Maximum size of a table.
-	kMaxTableSize = 2 * 1048576
-
-	// Maximum bytes of overlaps in grandparent (i.e., level+2) before we
-	// stop building a single file in a level->level+1 compaction.
-	kMaxGrandParentOverlapBytes = 10 * kMaxTableSize
-
-	// Maximum number of bytes in all compacted files.  We avoid expanding
-	// the lower level file set of a compaction if it would make the
-	// total compaction cover more than this many bytes.
-	kExpCompactionMaxBytes = 25 * kMaxTableSize
-)

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/db.go

@@ -7,15 +7,17 @@
 package leveldb
 
 import (
-	"errors"
+	"container/list"
 	"fmt"
 	"io"
 	"os"
 	"runtime"
 	"strings"
 	"sync"
+	"sync/atomic"
 	"time"
 
+	"github.com/syndtr/goleveldb/leveldb/errors"
 	"github.com/syndtr/goleveldb/leveldb/iterator"
 	"github.com/syndtr/goleveldb/leveldb/journal"
 	"github.com/syndtr/goleveldb/leveldb/memdb"
@@ -30,41 +32,46 @@ type DB struct {
 	// Need 64-bit alignment.
 	seq uint64
 
+	// Session.
 	s *session
 
-	// MemDB
+	// MemDB.
 	memMu             sync.RWMutex
-	mem               *memdb.DB
-	frozenMem         *memdb.DB
+	memPool           chan *memdb.DB
+	mem, frozenMem    *memDB
 	journal           *journal.Writer
 	journalWriter     storage.Writer
 	journalFile       storage.File
 	frozenJournalFile storage.File
 	frozenSeq         uint64
 
-	// Snapshot
+	// Snapshot.
 	snapsMu   sync.Mutex
-	snapsRoot snapshotElement
+	snapsList *list.List
 
-	// Write
+	// Stats.
+	aliveSnaps, aliveIters int32
+
+	// Write.
 	writeC       chan *Batch
 	writeMergedC chan bool
 	writeLockC   chan struct{}
 	writeAckC    chan error
+	writeDelay   time.Duration
+	writeDelayN  int
 	journalC     chan *Batch
 	journalAckC  chan error
 
-	// Compaction
-	tcompCmdC     chan cCmd
-	tcompPauseC   chan chan<- struct{}
-	tcompTriggerC chan struct{}
-	mcompCmdC     chan cCmd
-	mcompTriggerC chan struct{}
-	compErrC      chan error
-	compErrSetC   chan error
-	compStats     [kNumLevels]cStats
+	// Compaction.
+	tcompCmdC   chan cCmd
+	tcompPauseC chan chan<- struct{}
+	mcompCmdC   chan cCmd
+	compErrC    chan error
+	compPerErrC chan error
+	compErrSetC chan error
+	compStats   []cStats
 
-	// Close
+	// Close.
 	closeW sync.WaitGroup
 	closeC chan struct{}
 	closed uint32
@@ -77,7 +84,11 @@ func openDB(s *session) (*DB, error) {
 	db := &DB{
 		s: s,
 		// Initial sequence
-		seq: s.stSeq,
+		seq: s.stSeqNum,
+		// MemDB
+		memPool: make(chan *memdb.DB, 1),
+		// Snapshot
+		snapsList: list.New(),
 		// Write
 		writeC:       make(chan *Batch),
 		writeMergedC: make(chan bool),
@@ -86,17 +97,16 @@ func openDB(s *session) (*DB, error) {
 		journalC:     make(chan *Batch),
 		journalAckC:  make(chan error),
 		// Compaction
-		tcompCmdC:     make(chan cCmd),
-		tcompPauseC:   make(chan chan<- struct{}),
-		tcompTriggerC: make(chan struct{}, 1),
-		mcompCmdC:     make(chan cCmd),
-		mcompTriggerC: make(chan struct{}, 1),
-		compErrC:      make(chan error),
-		compErrSetC:   make(chan error),
+		tcompCmdC:   make(chan cCmd),
+		tcompPauseC: make(chan chan<- struct{}),
+		mcompCmdC:   make(chan cCmd),
+		compErrC:    make(chan error),
+		compPerErrC: make(chan error),
+		compErrSetC: make(chan error),
+		compStats:   make([]cStats, s.o.GetNumLevel()),
 		// Close
 		closeC: make(chan struct{}),
 	}
-	db.initSnapshot()
 
 	if err := db.recoverJournal(); err != nil {
 		return nil, err
@@ -112,8 +122,9 @@ func openDB(s *session) (*DB, error) {
 		return nil, err
 	}
 
-	// Don't include compaction error goroutine into wait group.
+	// Doesn't need to be included in the wait group.
 	go db.compactionError()
+	go db.mpoolDrain()
 
 	db.closeW.Add(3)
 	go db.tCompaction()
@@ -135,9 +146,10 @@ func openDB(s *session) (*DB, error) {
 // detected in the DB. Corrupted DB can be recovered with Recover
 // function.
 //
+// The returned DB instance is goroutine-safe.
 // The DB must be closed after use, by calling Close method.
-func Open(p storage.Storage, o *opt.Options) (db *DB, err error) {
-	s, err := newSession(p, o)
+func Open(stor storage.Storage, o *opt.Options) (db *DB, err error) {
+	s, err := newSession(stor, o)
 	if err != nil {
 		return
 	}
@@ -177,6 +189,7 @@ func Open(p storage.Storage, o *opt.Options) (db *DB, err error) {
 // detected in the DB. Corrupted DB can be recovered with Recover
 // function.
 //
+// The returned DB instance is goroutine-safe.
 // The DB must be closed after use, by calling Close method.
 func OpenFile(path string, o *opt.Options) (db *DB, err error) {
 	stor, err := storage.OpenFile(path)
@@ -197,9 +210,10 @@ func OpenFile(path string, o *opt.Options) (db *DB, err error) {
 // The DB must already exist or it will returns an error.
 // Also, Recover will ignore ErrorIfMissing and ErrorIfExist options.
 //
+// The returned DB instance is goroutine-safe.
 // The DB must be closed after use, by calling Close method.
-func Recover(p storage.Storage, o *opt.Options) (db *DB, err error) {
-	s, err := newSession(p, o)
+func Recover(stor storage.Storage, o *opt.Options) (db *DB, err error) {
+	s, err := newSession(stor, o)
 	if err != nil {
 		return
 	}
@@ -225,6 +239,7 @@ func Recover(p storage.Storage, o *opt.Options) (db *DB, err error) {
 // RecoverFile uses standard file-system backed storage implementation as desribed
 // in the leveldb/storage package.
 //
+// The returned DB instance is goroutine-safe.
 // The DB must be closed after use, by calling Close method.
 func RecoverFile(path string, o *opt.Options) (db *DB, err error) {
 	stor, err := storage.OpenFile(path)
@@ -241,16 +256,28 @@ func RecoverFile(path string, o *opt.Options) (db *DB, err error) {
 }
 
 func recoverTable(s *session, o *opt.Options) error {
-	ff0, err := s.getFiles(storage.TypeTable)
+	o = dupOptions(o)
+	// Mask StrictReader, lets StrictRecovery doing its job.
+	o.Strict &= ^opt.StrictReader
+
+	// Get all tables and sort it by file number.
+	tableFiles_, err := s.getFiles(storage.TypeTable)
 	if err != nil {
 		return err
 	}
-	ff1 := files(ff0)
-	ff1.sort()
+	tableFiles := files(tableFiles_)
+	tableFiles.sort()
 
-	var mSeq uint64
-	var good, corrupted int
-	rec := new(sessionRecord)
+	var (
+		mSeq                                                              uint64
+		recoveredKey, goodKey, corruptedKey, corruptedBlock, droppedTable int
+
+		// We will drop corrupted table.
+		strict = o.GetStrict(opt.StrictRecovery)
+
+		rec   = &sessionRecord{numLevel: o.GetNumLevel()}
+		bpool = util.NewBufferPool(o.GetBlockSize() + 5)
+	)
 	buildTable := func(iter iterator.Iterator) (tmp storage.File, size int64, err error) {
 		tmp = s.newTemp()
 		writer, err := tmp.Create()
@@ -264,8 +291,9 @@ func recoverTable(s *session, o *opt.Options) error {
 				tmp = nil
 			}
 		}()
+
+		// Copy entries.
 		tw := table.NewWriter(writer, o)
-		// Copy records.
 		for iter.Next() {
 			key := iter.Key()
 			if validIkey(key) {
@@ -297,44 +325,65 @@ func recoverTable(s *session, o *opt.Options) error {
 			return err
 		}
 		defer reader.Close()
+
 		// Get file size.
 		size, err := reader.Seek(0, 2)
 		if err != nil {
 			return err
 		}
-		var tSeq uint64
-		var tgood, tcorrupted, blockerr int
-		var min, max []byte
-		tr := table.NewReader(reader, size, nil, o)
+
+		var (
+			tSeq                                     uint64
+			tgoodKey, tcorruptedKey, tcorruptedBlock int
+			imin, imax                               []byte
+		)
+		tr, err := table.NewReader(reader, size, storage.NewFileInfo(file), nil, bpool, o)
+		if err != nil {
+			return err
+		}
 		iter := tr.NewIterator(nil, nil)
 		iter.(iterator.ErrorCallbackSetter).SetErrorCallback(func(err error) {
-			s.logf("table@recovery found error @%d %q", file.Num(), err)
-			blockerr++
+			if errors.IsCorrupted(err) {
+				s.logf("table@recovery block corruption @%d %q", file.Num(), err)
+				tcorruptedBlock++
+			}
 		})
+
 		// Scan the table.
 		for iter.Next() {
 			key := iter.Key()
-			_, seq, _, ok := parseIkey(key)
-			if !ok {
-				tcorrupted++
+			_, seq, _, kerr := parseIkey(key)
+			if kerr != nil {
+				tcorruptedKey++
 				continue
 			}
-			tgood++
+			tgoodKey++
 			if seq > tSeq {
 				tSeq = seq
 			}
-			if min == nil {
-				min = append([]byte{}, key...)
+			if imin == nil {
+				imin = append([]byte{}, key...)
 			}
-			max = append(max[:0], key...)
+			imax = append(imax[:0], key...)
 		}
 		if err := iter.Error(); err != nil {
 			iter.Release()
 			return err
 		}
 		iter.Release()
-		if tgood > 0 {
-			if tcorrupted > 0 || blockerr > 0 {
+
+		goodKey += tgoodKey
+		corruptedKey += tcorruptedKey
+		corruptedBlock += tcorruptedBlock
+
+		if strict && (tcorruptedKey > 0 || tcorruptedBlock > 0) {
+			droppedTable++
+			s.logf("table@recovery dropped @%d Gk·%d Ck·%d Cb·%d S·%d Q·%d", file.Num(), tgoodKey, tcorruptedKey, tcorruptedBlock, size, tSeq)
+			return nil
+		}
+
+		if tgoodKey > 0 {
+			if tcorruptedKey > 0 || tcorruptedBlock > 0 {
 				// Rebuild the table.
 				s.logf("table@recovery rebuilding @%d", file.Num())
 				iter := tr.NewIterator(nil, nil)
@@ -352,53 +401,67 @@ func recoverTable(s *session, o *opt.Options) error {
 			if tSeq > mSeq {
 				mSeq = tSeq
 			}
+			recoveredKey += tgoodKey
 			// Add table to level 0.
-			rec.addTable(0, file.Num(), uint64(size), min, max)
-			s.logf("table@recovery recovered @%d N·%d C·%d B·%d S·%d Q·%d", file.Num(), tgood, tcorrupted, blockerr, size, tSeq)
+			rec.addTable(0, file.Num(), uint64(size), imin, imax)
+			s.logf("table@recovery recovered @%d Gk·%d Ck·%d Cb·%d S·%d Q·%d", file.Num(), tgoodKey, tcorruptedKey, tcorruptedBlock, size, tSeq)
 		} else {
-			s.logf("table@recovery unrecoverable @%d C·%d B·%d S·%d", file.Num(), tcorrupted, blockerr, size)
+			droppedTable++
+			s.logf("table@recovery unrecoverable @%d Ck·%d Cb·%d S·%d", file.Num(), tcorruptedKey, tcorruptedBlock, size)
 		}
 
-		good += tgood
-		corrupted += tcorrupted
-
 		return nil
 	}
+
 	// Recover all tables.
-	if len(ff1) > 0 {
-		s.logf("table@recovery F·%d", len(ff1))
-		s.markFileNum(ff1[len(ff1)-1].Num())
-		for _, file := range ff1 {
+	if len(tableFiles) > 0 {
+		s.logf("table@recovery F·%d", len(tableFiles))
+
+		// Mark file number as used.
+		s.markFileNum(tableFiles[len(tableFiles)-1].Num())
+
+		for _, file := range tableFiles {
 			if err := recoverTable(file); err != nil {
 				return err
 			}
 		}
-		s.logf("table@recovery recovered F·%d N·%d C·%d Q·%d", len(ff1), good, corrupted, mSeq)
+
+		s.logf("table@recovery recovered F·%d N·%d Gk·%d Ck·%d Q·%d", len(tableFiles), recoveredKey, goodKey, corruptedKey, mSeq)
 	}
+
 	// Set sequence number.
-	rec.setSeq(mSeq + 1)
+	rec.setSeqNum(mSeq + 1)
+
 	// Create new manifest.
 	if err := s.create(); err != nil {
 		return err
 	}
+
 	// Commit.
 	return s.commit(rec)
 }
 
-func (d *DB) recoverJournal() error {
-	s := d.s
-
-	ff0, err := s.getFiles(storage.TypeJournal)
+func (db *DB) recoverJournal() error {
+	// Get all tables and sort it by file number.
+	journalFiles_, err := db.s.getFiles(storage.TypeJournal)
 	if err != nil {
 		return err
 	}
-	ff1 := files(ff0)
-	ff1.sort()
-	ff2 := make([]storage.File, 0, len(ff1))
-	for _, file := range ff1 {
-		if file.Num() >= s.stJournalNum || file.Num() == s.stPrevJournalNum {
-			s.markFileNum(file.Num())
-			ff2 = append(ff2, file)
+	journalFiles := files(journalFiles_)
+	journalFiles.sort()
+
+	// Discard older journal.
+	prev := -1
+	for i, file := range journalFiles {
+		if file.Num() >= db.s.stJournalNum {
+			if prev >= 0 {
+				i--
+				journalFiles[i] = journalFiles[prev]
+			}
+			journalFiles = journalFiles[i:]
+			break
+		} else if file.Num() == db.s.stPrevJournalNum {
+			prev = i
 		}
 	}
 
@@ -406,38 +469,43 @@ func (d *DB) recoverJournal() error {
 	var of storage.File
 	var mem *memdb.DB
 	batch := new(Batch)
-	cm := newCMem(s)
+	cm := newCMem(db.s)
 	buf := new(util.Buffer)
 	// Options.
-	strict := s.o.GetStrict(opt.StrictJournal)
-	checksum := s.o.GetStrict(opt.StrictJournalChecksum)
-	writeBuffer := s.o.GetWriteBuffer()
+	strict := db.s.o.GetStrict(opt.StrictJournal)
+	checksum := db.s.o.GetStrict(opt.StrictJournalChecksum)
+	writeBuffer := db.s.o.GetWriteBuffer()
 	recoverJournal := func(file storage.File) error {
-		s.logf("journal@recovery recovering @%d", file.Num())
+		db.logf("journal@recovery recovering @%d", file.Num())
 		reader, err := file.Open()
 		if err != nil {
 			return err
 		}
 		defer reader.Close()
+
+		// Create/reset journal reader instance.
 		if jr == nil {
-			jr = journal.NewReader(reader, dropper{s, file}, strict, checksum)
+			jr = journal.NewReader(reader, dropper{db.s, file}, strict, checksum)
 		} else {
-			jr.Reset(reader, dropper{s, file}, strict, checksum)
+			jr.Reset(reader, dropper{db.s, file}, strict, checksum)
 		}
+
+		// Flush memdb and remove obsolete journal file.
 		if of != nil {
 			if mem.Len() > 0 {
 				if err := cm.flush(mem, 0); err != nil {
 					return err
 				}
 			}
-			if err := cm.commit(file.Num(), d.seq); err != nil {
+			if err := cm.commit(file.Num(), db.seq); err != nil {
 				return err
 			}
 			cm.reset()
 			of.Remove()
 			of = nil
 		}
-		// Reset memdb.
+
+		// Replay journal to memdb.
 		mem.Reset()
 		for {
 			r, err := jr.Next()
@@ -445,43 +513,58 @@ func (d *DB) recoverJournal() error {
 				if err == io.EOF {
 					break
 				}
-				return err
+				return errors.SetFile(err, file)
 			}
+
 			buf.Reset()
 			if _, err := buf.ReadFrom(r); err != nil {
-				if strict {
-					return err
+				if err == io.ErrUnexpectedEOF {
+					// This is error returned due to corruption, with strict == false.
+					continue
+				} else {
+					return errors.SetFile(err, file)
 				}
-				continue
 			}
-			if err := batch.decode(buf.Bytes()); err != nil {
-				return err
+			if err := batch.memDecodeAndReplay(db.seq, buf.Bytes(), mem); err != nil {
+				if strict || !errors.IsCorrupted(err) {
+					return errors.SetFile(err, file)
+				} else {
+					db.s.logf("journal error: %v (skipped)", err)
+					// We won't apply sequence number as it might be corrupted.
+					continue
+				}
 			}
-			if err := batch.memReplay(mem); err != nil {
-				return err
-			}
-			d.seq = batch.seq + uint64(batch.len())
+
+			// Save sequence number.
+			db.seq = batch.seq + uint64(batch.Len())
+
+			// Flush it if large enough.
 			if mem.Size() >= writeBuffer {
-				// Large enough, flush it.
 				if err := cm.flush(mem, 0); err != nil {
 					return err
 				}
-				// Reset memdb.
 				mem.Reset()
 			}
 		}
+
 		of = file
 		return nil
 	}
+
 	// Recover all journals.
-	if len(ff2) > 0 {
-		s.logf("journal@recovery F·%d", len(ff2))
-		mem = memdb.New(s.icmp, writeBuffer)
-		for _, file := range ff2 {
+	if len(journalFiles) > 0 {
+		db.logf("journal@recovery F·%d", len(journalFiles))
+
+		// Mark file number as used.
+		db.s.markFileNum(journalFiles[len(journalFiles)-1].Num())
+
+		mem = memdb.New(db.s.icmp, writeBuffer)
+		for _, file := range journalFiles {
 			if err := recoverJournal(file); err != nil {
 				return err
 			}
 		}
+
 		// Flush the last journal.
 		if mem.Len() > 0 {
 			if err := cm.flush(mem, 0); err != nil {
@@ -489,51 +572,64 @@ func (d *DB) recoverJournal() error {
 			}
 		}
 	}
+
 	// Create a new journal.
-	if _, err := d.newMem(0); err != nil {
+	if _, err := db.newMem(0); err != nil {
 		return err
 	}
+
 	// Commit.
-	if err := cm.commit(d.journalFile.Num(), d.seq); err != nil {
+	if err := cm.commit(db.journalFile.Num(), db.seq); err != nil {
+		// Close journal.
+		if db.journal != nil {
+			db.journal.Close()
+			db.journalWriter.Close()
+		}
 		return err
 	}
-	// Remove the last journal.
+
+	// Remove the last obsolete journal file.
 	if of != nil {
 		of.Remove()
 	}
+
 	return nil
 }
 
-func (d *DB) get(key []byte, seq uint64, ro *opt.ReadOptions) (value []byte, err error) {
-	s := d.s
+func (db *DB) get(key []byte, seq uint64, ro *opt.ReadOptions) (value []byte, err error) {
+	ikey := newIkey(key, seq, ktSeek)
 
-	ikey := newIKey(key, seq, tSeek)
-
-	em, fm := d.getMems()
-	for _, m := range [...]*memdb.DB{em, fm} {
+	em, fm := db.getMems()
+	for _, m := range [...]*memDB{em, fm} {
 		if m == nil {
 			continue
 		}
-		mk, mv, me := m.Find(ikey)
+		defer m.decref()
+
+		mk, mv, me := m.mdb.Find(ikey)
 		if me == nil {
-			ukey, _, t, ok := parseIkey(mk)
-			if ok && s.icmp.uCompare(ukey, key) == 0 {
-				if t == tDel {
+			ukey, _, kt, kerr := parseIkey(mk)
+			if kerr != nil {
+				// Shouldn't have had happen.
+				panic(kerr)
+			}
+			if db.s.icmp.uCompare(ukey, key) == 0 {
+				if kt == ktDel {
 					return nil, ErrNotFound
 				}
-				return mv, nil
+				return append([]byte{}, mv...), nil
 			}
 		} else if me != ErrNotFound {
 			return nil, me
 		}
 	}
 
-	v := s.version()
+	v := db.s.version()
 	value, cSched, err := v.get(ikey, ro)
 	v.release()
 	if cSched {
 		// Trigger table compaction.
-		d.compTrigger(d.tcompTriggerC)
+		db.compSendTrigger(db.tcompCmdC)
 	}
 	return
 }
@@ -541,15 +637,18 @@ func (d *DB) get(key []byte, seq uint64, ro *opt.ReadOptions) (value []byte, err
 // Get gets the value for the given key. It returns ErrNotFound if the
 // DB does not contain the key.
 //
-// The caller should not modify the contents of the returned slice, but
-// it is safe to modify the contents of the argument after Get returns.
-func (d *DB) Get(key []byte, ro *opt.ReadOptions) (value []byte, err error) {
-	err = d.ok()
+// The returned slice is its own copy, it is safe to modify the contents
+// of the returned slice.
+// It is safe to modify the contents of the argument after Get returns.
+func (db *DB) Get(key []byte, ro *opt.ReadOptions) (value []byte, err error) {
+	err = db.ok()
 	if err != nil {
 		return
 	}
 
-	return d.get(key, d.getSeq(), ro)
+	se := db.acquireSnapshot()
+	defer db.releaseSnapshot(se)
+	return db.get(key, se.seq, ro)
 }
 
 // NewIterator returns an iterator for the latest snapshot of the
@@ -568,14 +667,16 @@ func (d *DB) Get(key []byte, ro *opt.ReadOptions) (value []byte, err error) {
 // The iterator must be released after use, by calling Release method.
 //
 // Also read Iterator documentation of the leveldb/iterator package.
-func (d *DB) NewIterator(slice *util.Range, ro *opt.ReadOptions) iterator.Iterator {
-	if err := d.ok(); err != nil {
+func (db *DB) NewIterator(slice *util.Range, ro *opt.ReadOptions) iterator.Iterator {
+	if err := db.ok(); err != nil {
 		return iterator.NewEmptyIterator(err)
 	}
 
-	p := d.newSnapshot()
-	defer p.Release()
-	return p.NewIterator(slice, ro)
+	se := db.acquireSnapshot()
+	defer db.releaseSnapshot(se)
+	// Iterator holds 'version' lock, 'version' is immutable so snapshot
+	// can be released after iterator created.
+	return db.newIterator(se.seq, slice, ro)
 }
 
 // GetSnapshot returns a latest snapshot of the underlying DB. A snapshot
@@ -583,25 +684,35 @@ func (d *DB) NewIterator(slice *util.Range, ro *opt.ReadOptions) iterator.Iterat
 // content of snapshot are guaranteed to be consistent.
 //
 // The snapshot must be released after use, by calling Release method.
-func (d *DB) GetSnapshot() (*Snapshot, error) {
-	if err := d.ok(); err != nil {
+func (db *DB) GetSnapshot() (*Snapshot, error) {
+	if err := db.ok(); err != nil {
 		return nil, err
 	}
 
-	return d.newSnapshot(), nil
+	return db.newSnapshot(), nil
 }
 
 // GetProperty returns value of the given property name.
 //
 // Property names:
 //	leveldb.num-files-at-level{n}
-//		Returns the number of filer at level 'n'.
+//		Returns the number of files at level 'n'.
 //	leveldb.stats
 //		Returns statistics of the underlying DB.
 //	leveldb.sstables
 //		Returns sstables list for each level.
-func (d *DB) GetProperty(name string) (value string, err error) {
-	err = d.ok()
+//	leveldb.blockpool
+//		Returns block pool stats.
+//	leveldb.cachedblock
+//		Returns size of cached block.
+//	leveldb.openedtables
+//		Returns number of opened tables.
+//	leveldb.alivesnaps
+//		Returns number of alive snapshots.
+//	leveldb.aliveiters
+//		Returns number of alive iterators.
+func (db *DB) GetProperty(name string) (value string, err error) {
+	err = db.ok()
 	if err != nil {
 		return
 	}
@@ -610,19 +721,18 @@ func (d *DB) GetProperty(name string) (value string, err error) {
 	if !strings.HasPrefix(name, prefix) {
 		return "", errors.New("leveldb: GetProperty: unknown property: " + name)
 	}
-
 	p := name[len(prefix):]
 
-	s := d.s
-	v := s.version()
+	v := db.s.version()
 	defer v.release()
 
+	numFilesPrefix := "num-files-at-level"
 	switch {
-	case strings.HasPrefix(p, "num-files-at-level"):
+	case strings.HasPrefix(p, numFilesPrefix):
 		var level uint
 		var rest string
-		n, _ := fmt.Scanf("%d%s", &level, &rest)
-		if n != 1 || level >= kNumLevels {
+		n, _ := fmt.Sscanf(p[len(numFilesPrefix):], "%d%s", &level, &rest)
+		if n != 1 || int(level) >= db.s.o.GetNumLevel() {
 			err = errors.New("leveldb: GetProperty: invalid property: " + name)
 		} else {
 			value = fmt.Sprint(v.tLen(int(level)))
@@ -631,22 +741,36 @@ func (d *DB) GetProperty(name string) (value string, err error) {
 		value = "Compactions\n" +
 			" Level |   Tables   |    Size(MB)   |    Time(sec)  |    Read(MB)   |   Write(MB)\n" +
 			"-------+------------+---------------+---------------+---------------+---------------\n"
-		for level, tt := range v.tables {
-			duration, read, write := d.compStats[level].get()
-			if len(tt) == 0 && duration == 0 {
+		for level, tables := range v.tables {
+			duration, read, write := db.compStats[level].get()
+			if len(tables) == 0 && duration == 0 {
 				continue
 			}
 			value += fmt.Sprintf(" %3d   | %10d | %13.5f | %13.5f | %13.5f | %13.5f\n",
-				level, len(tt), float64(tt.size())/1048576.0, duration.Seconds(),
+				level, len(tables), float64(tables.size())/1048576.0, duration.Seconds(),
 				float64(read)/1048576.0, float64(write)/1048576.0)
 		}
 	case p == "sstables":
-		for level, tt := range v.tables {
+		for level, tables := range v.tables {
 			value += fmt.Sprintf("--- level %d ---\n", level)
-			for _, t := range tt {
-				value += fmt.Sprintf("%d:%d[%q .. %q]\n", t.file.Num(), t.size, t.min, t.max)
+			for _, t := range tables {
+				value += fmt.Sprintf("%d:%d[%q .. %q]\n", t.file.Num(), t.size, t.imin, t.imax)
 			}
 		}
+	case p == "blockpool":
+		value = fmt.Sprintf("%v", db.s.tops.bpool)
+	case p == "cachedblock":
+		if bc := db.s.o.GetBlockCache(); bc != nil {
+			value = fmt.Sprintf("%d", bc.Size())
+		} else {
+			value = "<nil>"
+		}
+	case p == "openedtables":
+		value = fmt.Sprintf("%d", db.s.tops.cache.Size())
+	case p == "alivesnaps":
+		value = fmt.Sprintf("%d", atomic.LoadInt32(&db.aliveSnaps))
+	case p == "aliveiters":
+		value = fmt.Sprintf("%d", atomic.LoadInt32(&db.aliveIters))
 	default:
 		err = errors.New("leveldb: GetProperty: unknown property: " + name)
 	}
@@ -660,23 +784,23 @@ func (d *DB) GetProperty(name string) (value string, err error) {
 // data compresses by a factor of ten, the returned sizes will be one-tenth
 // the size of the corresponding user data size.
 // The results may not include the sizes of recently written data.
-func (d *DB) SizeOf(ranges []util.Range) (Sizes, error) {
-	if err := d.ok(); err != nil {
+func (db *DB) SizeOf(ranges []util.Range) (Sizes, error) {
+	if err := db.ok(); err != nil {
 		return nil, err
 	}
 
-	v := d.s.version()
+	v := db.s.version()
 	defer v.release()
 
 	sizes := make(Sizes, 0, len(ranges))
 	for _, r := range ranges {
-		min := newIKey(r.Start, kMaxSeq, tSeek)
-		max := newIKey(r.Limit, kMaxSeq, tSeek)
-		start, err := v.offsetOf(min)
+		imin := newIkey(r.Start, kMaxSeq, ktSeek)
+		imax := newIkey(r.Limit, kMaxSeq, ktSeek)
+		start, err := v.offsetOf(imin)
 		if err != nil {
 			return nil, err
 		}
-		limit, err := v.offsetOf(max)
+		limit, err := v.offsetOf(imax)
 		if err != nil {
 			return nil, err
 		}
@@ -690,61 +814,67 @@ func (d *DB) SizeOf(ranges []util.Range) (Sizes, error) {
 	return sizes, nil
 }
 
-// Close closes the DB. This will also releases any outstanding snapshot.
+// Close closes the DB. This will also releases any outstanding snapshot and
+// abort any in-flight compaction.
 //
 // It is not safe to close a DB until all outstanding iterators are released.
 // It is valid to call Close multiple times. Other methods should not be
 // called after the DB has been closed.
-func (d *DB) Close() error {
-	if !d.setClosed() {
+func (db *DB) Close() error {
+	if !db.setClosed() {
 		return ErrClosed
 	}
 
-	s := d.s
 	start := time.Now()
-	s.log("db@close closing")
+	db.log("db@close closing")
 
 	// Clear the finalizer.
-	runtime.SetFinalizer(d, nil)
+	runtime.SetFinalizer(db, nil)
 
 	// Get compaction error.
 	var err error
 	select {
-	case err = <-d.compErrC:
+	case err = <-db.compErrC:
 	default:
 	}
 
-	close(d.closeC)
+	// Signal all goroutines.
+	close(db.closeC)
 
-	// Wait for the close WaitGroup.
-	d.closeW.Wait()
+	// Wait for all gorotines to exit.
+	db.closeW.Wait()
 
-	// Close journal.
-	if d.journal != nil {
-		d.journal.Close()
-		d.journalWriter.Close()
+	// Lock writer and closes journal.
+	db.writeLockC <- struct{}{}
+	if db.journal != nil {
+		db.journal.Close()
+		db.journalWriter.Close()
+	}
+
+	if db.writeDelayN > 0 {
+		db.logf("db@write was delayed N·%d T·%v", db.writeDelayN, db.writeDelay)
 	}
 
 	// Close session.
-	s.close()
-	s.logf("db@close done T·%v", time.Since(start))
-	s.release()
+	db.s.close()
+	db.logf("db@close done T·%v", time.Since(start))
+	db.s.release()
 
-	if d.closer != nil {
-		if err1 := d.closer.Close(); err == nil {
+	if db.closer != nil {
+		if err1 := db.closer.Close(); err == nil {
 			err = err1
 		}
 	}
 
-	d.s = nil
-	d.mem = nil
-	d.frozenMem = nil
-	d.journal = nil
-	d.journalWriter = nil
-	d.journalFile = nil
-	d.frozenJournalFile = nil
-	d.snapsRoot = snapshotElement{}
-	d.closer = nil
+	// NIL'ing pointers.
+	db.s = nil
+	db.mem = nil
+	db.frozenMem = nil
+	db.journal = nil
+	db.journalWriter = nil
+	db.journalFile = nil
+	db.frozenJournalFile = nil
+	db.closer = nil
 
 	return err
 }

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/db_iter.go

@@ -9,6 +9,8 @@ package leveldb
 import (
 	"errors"
 	"runtime"
+	"sync"
+	"sync/atomic"
 
 	"github.com/syndtr/goleveldb/leveldb/iterator"
 	"github.com/syndtr/goleveldb/leveldb/opt"
@@ -19,46 +21,61 @@ var (
 	errInvalidIkey = errors.New("leveldb: Iterator: invalid internal key")
 )
 
-func (db *DB) newRawIterator(slice *util.Range, ro *opt.ReadOptions) iterator.Iterator {
-	s := db.s
+type memdbReleaser struct {
+	once sync.Once
+	m    *memDB
+}
 
+func (mr *memdbReleaser) Release() {
+	mr.once.Do(func() {
+		mr.m.decref()
+	})
+}
+
+func (db *DB) newRawIterator(slice *util.Range, ro *opt.ReadOptions) iterator.Iterator {
 	em, fm := db.getMems()
-	v := s.version()
+	v := db.s.version()
 
 	ti := v.getIterators(slice, ro)
 	n := len(ti) + 2
 	i := make([]iterator.Iterator, 0, n)
-	i = append(i, em.NewIterator(slice))
+	emi := em.mdb.NewIterator(slice)
+	emi.SetReleaser(&memdbReleaser{m: em})
+	i = append(i, emi)
 	if fm != nil {
-		i = append(i, fm.NewIterator(slice))
+		fmi := fm.mdb.NewIterator(slice)
+		fmi.SetReleaser(&memdbReleaser{m: fm})
+		i = append(i, fmi)
 	}
 	i = append(i, ti...)
-	strict := s.o.GetStrict(opt.StrictIterator) || ro.GetStrict(opt.StrictIterator)
-	mi := iterator.NewMergedIterator(i, s.icmp, strict)
+	strict := opt.GetStrict(db.s.o.Options, ro, opt.StrictReader)
+	mi := iterator.NewMergedIterator(i, db.s.icmp, strict)
 	mi.SetReleaser(&versionReleaser{v: v})
 	return mi
 }
 
 func (db *DB) newIterator(seq uint64, slice *util.Range, ro *opt.ReadOptions) *dbIter {
-	var slice_ *util.Range
+	var islice *util.Range
 	if slice != nil {
-		slice_ = &util.Range{}
+		islice = &util.Range{}
 		if slice.Start != nil {
-			slice_.Start = newIKey(slice.Start, kMaxSeq, tSeek)
+			islice.Start = newIkey(slice.Start, kMaxSeq, ktSeek)
 		}
 		if slice.Limit != nil {
-			slice_.Limit = newIKey(slice.Limit, kMaxSeq, tSeek)
+			islice.Limit = newIkey(slice.Limit, kMaxSeq, ktSeek)
 		}
 	}
-	rawIter := db.newRawIterator(slice_, ro)
+	rawIter := db.newRawIterator(islice, ro)
 	iter := &dbIter{
+		db:     db,
 		icmp:   db.s.icmp,
 		iter:   rawIter,
 		seq:    seq,
-		strict: db.s.o.GetStrict(opt.StrictIterator) || ro.GetStrict(opt.StrictIterator),
+		strict: opt.GetStrict(db.s.o.Options, ro, opt.StrictReader),
 		key:    make([]byte, 0),
 		value:  make([]byte, 0),
 	}
+	atomic.AddInt32(&db.aliveIters, 1)
 	runtime.SetFinalizer(iter, (*dbIter).Release)
 	return iter
 }
@@ -75,6 +92,7 @@ const (
 
 // dbIter represent an interator states over a database session.
 type dbIter struct {
+	db     *DB
 	icmp   *iComparer
 	iter   iterator.Iterator
 	seq    uint64
@@ -144,7 +162,7 @@ func (i *dbIter) Seek(key []byte) bool {
 		return false
 	}
 
-	ikey := newIKey(key, i.seq, tSeek)
+	ikey := newIkey(key, i.seq, ktSeek)
 	if i.iter.Seek(ikey) {
 		i.dir = dirSOI
 		return i.next()
@@ -156,15 +174,14 @@ func (i *dbIter) Seek(key []byte) bool {
 
 func (i *dbIter) next() bool {
 	for {
-		ukey, seq, t, ok := parseIkey(i.iter.Key())
-		if ok {
+		if ukey, seq, kt, kerr := parseIkey(i.iter.Key()); kerr == nil {
 			if seq <= i.seq {
-				switch t {
-				case tDel:
+				switch kt {
+				case ktDel:
 					// Skip deleted key.
 					i.key = append(i.key[:0], ukey...)
 					i.dir = dirForward
-				case tVal:
+				case ktVal:
 					if i.dir == dirSOI || i.icmp.uCompare(ukey, i.key) > 0 {
 						i.key = append(i.key[:0], ukey...)
 						i.value = append(i.value[:0], i.iter.Value()...)
@@ -174,7 +191,7 @@ func (i *dbIter) next() bool {
 				}
 			}
 		} else if i.strict {
-			i.setErr(errInvalidIkey)
+			i.setErr(kerr)
 			break
 		}
 		if !i.iter.Next() {
@@ -207,20 +224,19 @@ func (i *dbIter) prev() bool {
 	del := true
 	if i.iter.Valid() {
 		for {
-			ukey, seq, t, ok := parseIkey(i.iter.Key())
-			if ok {
+			if ukey, seq, kt, kerr := parseIkey(i.iter.Key()); kerr == nil {
 				if seq <= i.seq {
 					if !del && i.icmp.uCompare(ukey, i.key) < 0 {
 						return true
 					}
-					del = (t == tDel)
+					del = (kt == ktDel)
 					if !del {
 						i.key = append(i.key[:0], ukey...)
 						i.value = append(i.value[:0], i.iter.Value()...)
 					}
 				}
 			} else if i.strict {
-				i.setErr(errInvalidIkey)
+				i.setErr(kerr)
 				return false
 			}
 			if !i.iter.Prev() {
@@ -249,13 +265,12 @@ func (i *dbIter) Prev() bool {
 		return i.Last()
 	case dirForward:
 		for i.iter.Prev() {
-			ukey, _, _, ok := parseIkey(i.iter.Key())
-			if ok {
+			if ukey, _, _, kerr := parseIkey(i.iter.Key()); kerr == nil {
 				if i.icmp.uCompare(ukey, i.key) < 0 {
 					goto cont
 				}
 			} else if i.strict {
-				i.setErr(errInvalidIkey)
+				i.setErr(kerr)
 				return false
 			}
 		}
@@ -289,6 +304,7 @@ func (i *dbIter) Release() {
 
 		if i.releaser != nil {
 			i.releaser.Release()
+			i.releaser = nil
 		}
 
 		i.dir = dirReleased
@@ -296,13 +312,19 @@ func (i *dbIter) Release() {
 		i.value = nil
 		i.iter.Release()
 		i.iter = nil
+		atomic.AddInt32(&i.db.aliveIters, -1)
+		i.db = nil
 	}
 }
 
 func (i *dbIter) SetReleaser(releaser util.Releaser) {
-	if i.dir != dirReleased {
-		i.releaser = releaser
+	if i.dir == dirReleased {
+		panic(util.ErrReleased)
 	}
+	if i.releaser != nil && releaser != nil {
+		panic(util.ErrHasReleaser)
+	}
+	i.releaser = releaser
 }
 
 func (i *dbIter) Error() error {

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/db_snapshot.go

@@ -7,8 +7,10 @@
 package leveldb
 
 import (
+	"container/list"
 	"runtime"
 	"sync"
+	"sync/atomic"
 
 	"github.com/syndtr/goleveldb/leveldb/iterator"
 	"github.com/syndtr/goleveldb/leveldb/opt"
@@ -18,51 +20,41 @@ import (
 type snapshotElement struct {
 	seq uint64
 	ref int
-	// Next and previous pointers in the doubly-linked list of elements.
-	next, prev *snapshotElement
-}
-
-// Initialize the snapshot.
-func (db *DB) initSnapshot() {
-	db.snapsRoot.next = &db.snapsRoot
-	db.snapsRoot.prev = &db.snapsRoot
+	e   *list.Element
 }
 
 // Acquires a snapshot, based on latest sequence.
 func (db *DB) acquireSnapshot() *snapshotElement {
 	db.snapsMu.Lock()
+	defer db.snapsMu.Unlock()
+
 	seq := db.getSeq()
-	elem := db.snapsRoot.prev
-	if elem == &db.snapsRoot || elem.seq != seq {
-		at := db.snapsRoot.prev
-		next := at.next
-		elem = &snapshotElement{
-			seq:  seq,
-			prev: at,
-			next: next,
+
+	if e := db.snapsList.Back(); e != nil {
+		se := e.Value.(*snapshotElement)
+		if se.seq == seq {
+			se.ref++
+			return se
+		} else if seq < se.seq {
+			panic("leveldb: sequence number is not increasing")
 		}
-		at.next = elem
-		next.prev = elem
 	}
-	elem.ref++
-	db.snapsMu.Unlock()
-	return elem
+	se := &snapshotElement{seq: seq, ref: 1}
+	se.e = db.snapsList.PushBack(se)
+	return se
 }
 
 // Releases given snapshot element.
-func (db *DB) releaseSnapshot(elem *snapshotElement) {
-	if !db.isClosed() {
-		db.snapsMu.Lock()
-		elem.ref--
-		if elem.ref == 0 {
-			elem.prev.next = elem.next
-			elem.next.prev = elem.prev
-			elem.next = nil
-			elem.prev = nil
-		} else if elem.ref < 0 {
-			panic("leveldb: Snapshot: negative element reference")
-		}
-		db.snapsMu.Unlock()
+func (db *DB) releaseSnapshot(se *snapshotElement) {
+	db.snapsMu.Lock()
+	defer db.snapsMu.Unlock()
+
+	se.ref--
+	if se.ref == 0 {
+		db.snapsList.Remove(se.e)
+		se.e = nil
+	} else if se.ref < 0 {
+		panic("leveldb: Snapshot: negative element reference")
 	}
 }
 
@@ -70,10 +62,11 @@ func (db *DB) releaseSnapshot(elem *snapshotElement) {
 func (db *DB) minSeq() uint64 {
 	db.snapsMu.Lock()
 	defer db.snapsMu.Unlock()
-	elem := db.snapsRoot.prev
-	if elem != &db.snapsRoot {
-		return elem.seq
+
+	if e := db.snapsList.Front(); e != nil {
+		return e.Value.(*snapshotElement).seq
 	}
+
 	return db.getSeq()
 }
 
@@ -81,18 +74,19 @@ func (db *DB) minSeq() uint64 {
 type Snapshot struct {
 	db       *DB
 	elem     *snapshotElement
-	mu       sync.Mutex
+	mu       sync.RWMutex
 	released bool
 }
 
 // Creates new snapshot object.
 func (db *DB) newSnapshot() *Snapshot {
-	p := &Snapshot{
+	snap := &Snapshot{
 		db:   db,
 		elem: db.acquireSnapshot(),
 	}
-	runtime.SetFinalizer(p, (*Snapshot).Release)
-	return p
+	atomic.AddInt32(&db.aliveSnaps, 1)
+	runtime.SetFinalizer(snap, (*Snapshot).Release)
+	return snap
 }
 
 // Get gets the value for the given key. It returns ErrNotFound if
@@ -100,19 +94,18 @@ func (db *DB) newSnapshot() *Snapshot {
 //
 // The caller should not modify the contents of the returned slice, but
 // it is safe to modify the contents of the argument after Get returns.
-func (p *Snapshot) Get(key []byte, ro *opt.ReadOptions) (value []byte, err error) {
-	db := p.db
-	err = db.ok()
+func (snap *Snapshot) Get(key []byte, ro *opt.ReadOptions) (value []byte, err error) {
+	err = snap.db.ok()
 	if err != nil {
 		return
 	}
-	p.mu.Lock()
-	defer p.mu.Unlock()
-	if p.released {
+	snap.mu.RLock()
+	defer snap.mu.RUnlock()
+	if snap.released {
 		err = ErrSnapshotReleased
 		return
 	}
-	return db.get(key, p.elem.seq, ro)
+	return snap.db.get(key, snap.elem.seq, ro)
 }
 
 // NewIterator returns an iterator for the snapshot of the uderlying DB.
@@ -132,17 +125,18 @@ func (p *Snapshot) Get(key []byte, ro *opt.ReadOptions) (value []byte, err error
 // iterator would be still valid until released.
 //
 // Also read Iterator documentation of the leveldb/iterator package.
-func (p *Snapshot) NewIterator(slice *util.Range, ro *opt.ReadOptions) iterator.Iterator {
-	db := p.db
-	if err := db.ok(); err != nil {
+func (snap *Snapshot) NewIterator(slice *util.Range, ro *opt.ReadOptions) iterator.Iterator {
+	if err := snap.db.ok(); err != nil {
 		return iterator.NewEmptyIterator(err)
 	}
-	p.mu.Lock()
-	defer p.mu.Unlock()
-	if p.released {
+	snap.mu.Lock()
+	defer snap.mu.Unlock()
+	if snap.released {
 		return iterator.NewEmptyIterator(ErrSnapshotReleased)
 	}
-	return db.newIterator(p.elem.seq, slice, ro)
+	// Since iterator already hold version ref, it doesn't need to
+	// hold snapshot ref.
+	return snap.db.newIterator(snap.elem.seq, slice, ro)
 }
 
 // Release releases the snapshot. This will not release any returned
@@ -150,16 +144,18 @@ func (p *Snapshot) NewIterator(slice *util.Range, ro *opt.ReadOptions) iterator.
 // underlying DB is closed.
 //
 // Other methods should not be called after the snapshot has been released.
-func (p *Snapshot) Release() {
-	p.mu.Lock()
-	if !p.released {
-		// Clear the finalizer.
-		runtime.SetFinalizer(p, nil)
+func (snap *Snapshot) Release() {
+	snap.mu.Lock()
+	defer snap.mu.Unlock()
 
-		p.released = true
-		p.db.releaseSnapshot(p.elem)
-		p.db = nil
-		p.elem = nil
+	if !snap.released {
+		// Clear the finalizer.
+		runtime.SetFinalizer(snap, nil)
+
+		snap.released = true
+		snap.db.releaseSnapshot(snap.elem)
+		atomic.AddInt32(&snap.db.aliveSnaps, -1)
+		snap.db = nil
+		snap.elem = nil
 	}
-	p.mu.Unlock()
 }

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/db_state.go

@@ -8,106 +8,194 @@ package leveldb
 
 import (
 	"sync/atomic"
+	"time"
 
 	"github.com/syndtr/goleveldb/leveldb/journal"
 	"github.com/syndtr/goleveldb/leveldb/memdb"
 )
 
+type memDB struct {
+	db  *DB
+	mdb *memdb.DB
+	ref int32
+}
+
+func (m *memDB) incref() {
+	atomic.AddInt32(&m.ref, 1)
+}
+
+func (m *memDB) decref() {
+	if ref := atomic.AddInt32(&m.ref, -1); ref == 0 {
+		// Only put back memdb with std capacity.
+		if m.mdb.Capacity() == m.db.s.o.GetWriteBuffer() {
+			m.mdb.Reset()
+			m.db.mpoolPut(m.mdb)
+		}
+		m.db = nil
+		m.mdb = nil
+	} else if ref < 0 {
+		panic("negative memdb ref")
+	}
+}
+
 // Get latest sequence number.
-func (d *DB) getSeq() uint64 {
-	return atomic.LoadUint64(&d.seq)
+func (db *DB) getSeq() uint64 {
+	return atomic.LoadUint64(&db.seq)
 }
 
 // Atomically adds delta to seq.
-func (d *DB) addSeq(delta uint64) {
-	atomic.AddUint64(&d.seq, delta)
+func (db *DB) addSeq(delta uint64) {
+	atomic.AddUint64(&db.seq, delta)
+}
+
+func (db *DB) mpoolPut(mem *memdb.DB) {
+	defer func() {
+		recover()
+	}()
+	select {
+	case db.memPool <- mem:
+	default:
+	}
+}
+
+func (db *DB) mpoolGet() *memdb.DB {
+	select {
+	case mem := <-db.memPool:
+		return mem
+	default:
+		return nil
+	}
+}
+
+func (db *DB) mpoolDrain() {
+	ticker := time.NewTicker(30 * time.Second)
+	for {
+		select {
+		case <-ticker.C:
+			select {
+			case <-db.memPool:
+			default:
+			}
+		case _, _ = <-db.closeC:
+			close(db.memPool)
+			return
+		}
+	}
 }
 
 // Create new memdb and froze the old one; need external synchronization.
 // newMem only called synchronously by the writer.
-func (d *DB) newMem(n int) (mem *memdb.DB, err error) {
-	s := d.s
-
-	num := s.allocFileNum()
-	file := s.getJournalFile(num)
+func (db *DB) newMem(n int) (mem *memDB, err error) {
+	num := db.s.allocFileNum()
+	file := db.s.getJournalFile(num)
 	w, err := file.Create()
 	if err != nil {
-		s.reuseFileNum(num)
+		db.s.reuseFileNum(num)
 		return
 	}
-	d.memMu.Lock()
-	if d.journal == nil {
-		d.journal = journal.NewWriter(w)
-	} else {
-		d.journal.Reset(w)
-		d.journalWriter.Close()
-		d.frozenJournalFile = d.journalFile
+
+	db.memMu.Lock()
+	defer db.memMu.Unlock()
+
+	if db.frozenMem != nil {
+		panic("still has frozen mem")
 	}
-	d.journalWriter = w
-	d.journalFile = file
-	d.frozenMem = d.mem
-	d.mem = memdb.New(s.icmp, maxInt(d.s.o.GetWriteBuffer(), n))
-	mem = d.mem
-	// The seq only incremented by the writer.
-	d.frozenSeq = d.seq
-	d.memMu.Unlock()
+
+	if db.journal == nil {
+		db.journal = journal.NewWriter(w)
+	} else {
+		db.journal.Reset(w)
+		db.journalWriter.Close()
+		db.frozenJournalFile = db.journalFile
+	}
+	db.journalWriter = w
+	db.journalFile = file
+	db.frozenMem = db.mem
+	mdb := db.mpoolGet()
+	if mdb == nil || mdb.Capacity() < n {
+		mdb = memdb.New(db.s.icmp, maxInt(db.s.o.GetWriteBuffer(), n))
+	}
+	mem = &memDB{
+		db:  db,
+		mdb: mdb,
+		ref: 2,
+	}
+	db.mem = mem
+	// The seq only incremented by the writer. And whoever called newMem
+	// should hold write lock, so no need additional synchronization here.
+	db.frozenSeq = db.seq
 	return
 }
 
 // Get all memdbs.
-func (d *DB) getMems() (e *memdb.DB, f *memdb.DB) {
-	d.memMu.RLock()
-	defer d.memMu.RUnlock()
-	return d.mem, d.frozenMem
+func (db *DB) getMems() (e, f *memDB) {
+	db.memMu.RLock()
+	defer db.memMu.RUnlock()
+	if db.mem == nil {
+		panic("nil effective mem")
+	}
+	db.mem.incref()
+	if db.frozenMem != nil {
+		db.frozenMem.incref()
+	}
+	return db.mem, db.frozenMem
 }
 
 // Get frozen memdb.
-func (d *DB) getEffectiveMem() *memdb.DB {
-	d.memMu.RLock()
-	defer d.memMu.RUnlock()
-	return d.mem
+func (db *DB) getEffectiveMem() *memDB {
+	db.memMu.RLock()
+	defer db.memMu.RUnlock()
+	if db.mem == nil {
+		panic("nil effective mem")
+	}
+	db.mem.incref()
+	return db.mem
 }
 
 // Check whether we has frozen memdb.
-func (d *DB) hasFrozenMem() bool {
-	d.memMu.RLock()
-	defer d.memMu.RUnlock()
-	return d.frozenMem != nil
+func (db *DB) hasFrozenMem() bool {
+	db.memMu.RLock()
+	defer db.memMu.RUnlock()
+	return db.frozenMem != nil
 }
 
 // Get frozen memdb.
-func (d *DB) getFrozenMem() *memdb.DB {
-	d.memMu.RLock()
-	defer d.memMu.RUnlock()
-	return d.frozenMem
+func (db *DB) getFrozenMem() *memDB {
+	db.memMu.RLock()
+	defer db.memMu.RUnlock()
+	if db.frozenMem != nil {
+		db.frozenMem.incref()
+	}
+	return db.frozenMem
 }
 
 // Drop frozen memdb; assume that frozen memdb isn't nil.
-func (d *DB) dropFrozenMem() {
-	d.memMu.Lock()
-	if err := d.frozenJournalFile.Remove(); err != nil {
-		d.s.logf("journal@remove removing @%d %q", d.frozenJournalFile.Num(), err)
+func (db *DB) dropFrozenMem() {
+	db.memMu.Lock()
+	if err := db.frozenJournalFile.Remove(); err != nil {
+		db.logf("journal@remove removing @%d %q", db.frozenJournalFile.Num(), err)
 	} else {
-		d.s.logf("journal@remove removed @%d", d.frozenJournalFile.Num())
+		db.logf("journal@remove removed @%d", db.frozenJournalFile.Num())
 	}
-	d.frozenJournalFile = nil
-	d.frozenMem = nil
-	d.memMu.Unlock()
+	db.frozenJournalFile = nil
+	db.frozenMem.decref()
+	db.frozenMem = nil
+	db.memMu.Unlock()
 }
 
 // Set closed flag; return true if not already closed.
-func (d *DB) setClosed() bool {
-	return atomic.CompareAndSwapUint32(&d.closed, 0, 1)
+func (db *DB) setClosed() bool {
+	return atomic.CompareAndSwapUint32(&db.closed, 0, 1)
 }
 
 // Check whether DB was closed.
-func (d *DB) isClosed() bool {
-	return atomic.LoadUint32(&d.closed) != 0
+func (db *DB) isClosed() bool {
+	return atomic.LoadUint32(&db.closed) != 0
 }
 
 // Check read ok status.
-func (d *DB) ok() error {
-	if d.isClosed() {
+func (db *DB) ok() error {
+	if db.isClosed() {
 		return ErrClosed
 	}
 	return nil

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/db_test.go

@@ -7,6 +7,10 @@
 package leveldb
 
 import (
+	"bytes"
+	"container/list"
+	crand "crypto/rand"
+	"encoding/binary"
 	"fmt"
 	"math/rand"
 	"os"
@@ -20,6 +24,7 @@ import (
 	"unsafe"
 
 	"github.com/syndtr/goleveldb/leveldb/comparer"
+	"github.com/syndtr/goleveldb/leveldb/errors"
 	"github.com/syndtr/goleveldb/leveldb/filter"
 	"github.com/syndtr/goleveldb/leveldb/iterator"
 	"github.com/syndtr/goleveldb/leveldb/opt"
@@ -148,25 +153,29 @@ func (h *dbHarness) maxNextLevelOverlappingBytes(want uint64) {
 	t := h.t
 	db := h.db
 
-	var res uint64
+	var (
+		maxOverlaps uint64
+		maxLevel    int
+	)
 	v := db.s.version()
 	for i, tt := range v.tables[1 : len(v.tables)-1] {
 		level := i + 1
 		next := v.tables[level+1]
 		for _, t := range tt {
-			var r tFiles
-			min, max := t.min.ukey(), t.max.ukey()
-			next.getOverlaps(min, max, &r, true, db.s.icmp.ucmp)
+			r := next.getOverlaps(nil, db.s.icmp, t.imin.ukey(), t.imax.ukey(), false)
 			sum := r.size()
-			if sum > res {
-				res = sum
+			if sum > maxOverlaps {
+				maxOverlaps = sum
+				maxLevel = level
 			}
 		}
 	}
 	v.release()
 
-	if res > want {
-		t.Errorf("next level overlapping bytes is more than %d, got=%d", want, res)
+	if maxOverlaps > want {
+		t.Errorf("next level most overlapping bytes is more than %d, got=%d level=%d", want, maxOverlaps, maxLevel)
+	} else {
+		t.Logf("next level most overlapping bytes is %d, level=%d want=%d", maxOverlaps, maxLevel, want)
 	}
 }
 
@@ -239,7 +248,7 @@ func (h *dbHarness) allEntriesFor(key, want string) {
 	db := h.db
 	s := db.s
 
-	ikey := newIKey([]byte(key), kMaxSeq, tVal)
+	ikey := newIkey([]byte(key), kMaxSeq, ktVal)
 	iter := db.newRawIterator(nil, nil)
 	if !iter.Seek(ikey) && iter.Error() != nil {
 		t.Error("AllEntries: error during seek, err: ", iter.Error())
@@ -248,19 +257,18 @@ func (h *dbHarness) allEntriesFor(key, want string) {
 	res := "[ "
 	first := true
 	for iter.Valid() {
-		rkey := iKey(iter.Key())
-		if _, t, ok := rkey.parseNum(); ok {
-			if s.icmp.uCompare(ikey.ukey(), rkey.ukey()) != 0 {
+		if ukey, _, kt, kerr := parseIkey(iter.Key()); kerr == nil {
+			if s.icmp.uCompare(ikey.ukey(), ukey) != 0 {
 				break
 			}
 			if !first {
 				res += ", "
 			}
 			first = false
-			switch t {
-			case tVal:
+			switch kt {
+			case ktVal:
 				res += string(iter.Value())
-			case tDel:
+			case ktDel:
 				res += "DEL"
 			}
 		} else {
@@ -325,6 +333,8 @@ func (h *dbHarness) compactMem() {
 	t := h.t
 	db := h.db
 
+	t.Log("starting memdb compaction")
+
 	db.writeLockC <- struct{}{}
 	defer func() {
 		<-db.writeLockC
@@ -340,6 +350,8 @@ func (h *dbHarness) compactMem() {
 	if h.totalTables() == 0 {
 		t.Error("zero tables after mem compaction")
 	}
+
+	t.Log("memdb compaction done")
 }
 
 func (h *dbHarness) compactRangeAtErr(level int, min, max string, wanterr bool) {
@@ -354,6 +366,8 @@ func (h *dbHarness) compactRangeAtErr(level int, min, max string, wanterr bool)
 		_max = []byte(max)
 	}
 
+	t.Logf("starting table range compaction: level=%d, min=%q, max=%q", level, min, max)
+
 	if err := db.compSendRange(db.tcompCmdC, level, _min, _max); err != nil {
 		if wanterr {
 			t.Log("CompactRangeAt: got error (expected): ", err)
@@ -363,6 +377,8 @@ func (h *dbHarness) compactRangeAtErr(level int, min, max string, wanterr bool)
 	} else if wanterr {
 		t.Error("CompactRangeAt: expect error")
 	}
+
+	t.Log("table range compaction done")
 }
 
 func (h *dbHarness) compactRangeAt(level int, min, max string) {
@@ -373,6 +389,8 @@ func (h *dbHarness) compactRange(min, max string) {
 	t := h.t
 	db := h.db
 
+	t.Logf("starting DB range compaction: min=%q, max=%q", min, max)
+
 	var r util.Range
 	if min != "" {
 		r.Start = []byte(min)
@@ -383,6 +401,8 @@ func (h *dbHarness) compactRange(min, max string) {
 	if err := db.CompactRange(r); err != nil {
 		t.Error("CompactRange: got error: ", err)
 	}
+
+	t.Log("DB range compaction done")
 }
 
 func (h *dbHarness) sizeAssert(start, limit string, low, hi uint64) {
@@ -504,10 +524,10 @@ func Test_FieldsAligned(t *testing.T) {
 	p1 := new(DB)
 	testAligned(t, "DB.seq", unsafe.Offsetof(p1.seq))
 	p2 := new(session)
-	testAligned(t, "session.stFileNum", unsafe.Offsetof(p2.stFileNum))
+	testAligned(t, "session.stNextFileNum", unsafe.Offsetof(p2.stNextFileNum))
 	testAligned(t, "session.stJournalNum", unsafe.Offsetof(p2.stJournalNum))
 	testAligned(t, "session.stPrevJournalNum", unsafe.Offsetof(p2.stPrevJournalNum))
-	testAligned(t, "session.stSeq", unsafe.Offsetof(p2.stSeq))
+	testAligned(t, "session.stSeqNum", unsafe.Offsetof(p2.stSeqNum))
 }
 
 func TestDb_Locking(t *testing.T) {
@@ -943,7 +963,7 @@ func TestDb_RepeatedWritesToSameKey(t *testing.T) {
 	h := newDbHarnessWopt(t, &opt.Options{WriteBuffer: 100000})
 	defer h.close()
 
-	maxTables := kNumLevels + kL0_StopWritesTrigger
+	maxTables := h.o.GetNumLevel() + h.o.GetWriteL0PauseTrigger()
 
 	value := strings.Repeat("v", 2*h.o.GetWriteBuffer())
 	for i := 0; i < 5*maxTables; i++ {
@@ -961,7 +981,7 @@ func TestDb_RepeatedWritesToSameKeyAfterReopen(t *testing.T) {
 
 	h.reopenDB()
 
-	maxTables := kNumLevels + kL0_StopWritesTrigger
+	maxTables := h.o.GetNumLevel() + h.o.GetWriteL0PauseTrigger()
 
 	value := strings.Repeat("v", 2*h.o.GetWriteBuffer())
 	for i := 0; i < 5*maxTables; i++ {
@@ -977,7 +997,7 @@ func TestDb_SparseMerge(t *testing.T) {
 	h := newDbHarnessWopt(t, &opt.Options{Compression: opt.NoCompression})
 	defer h.close()
 
-	h.putMulti(kNumLevels, "A", "Z")
+	h.putMulti(h.o.GetNumLevel(), "A", "Z")
 
 	// Suppose there is:
 	//    small amount of data with prefix A
@@ -1001,6 +1021,7 @@ func TestDb_SparseMerge(t *testing.T) {
 	h.put("C", "vc2")
 	h.compactMem()
 
+	h.waitCompaction()
 	h.maxNextLevelOverlappingBytes(20 * 1048576)
 	h.compactRangeAt(0, "", "")
 	h.waitCompaction()
@@ -1127,13 +1148,51 @@ func TestDb_Snapshot(t *testing.T) {
 	})
 }
 
+func TestDb_SnapshotList(t *testing.T) {
+	db := &DB{snapsList: list.New()}
+	e0a := db.acquireSnapshot()
+	e0b := db.acquireSnapshot()
+	db.seq = 1
+	e1 := db.acquireSnapshot()
+	db.seq = 2
+	e2 := db.acquireSnapshot()
+
+	if db.minSeq() != 0 {
+		t.Fatalf("invalid sequence number, got=%d", db.minSeq())
+	}
+	db.releaseSnapshot(e0a)
+	if db.minSeq() != 0 {
+		t.Fatalf("invalid sequence number, got=%d", db.minSeq())
+	}
+	db.releaseSnapshot(e2)
+	if db.minSeq() != 0 {
+		t.Fatalf("invalid sequence number, got=%d", db.minSeq())
+	}
+	db.releaseSnapshot(e0b)
+	if db.minSeq() != 1 {
+		t.Fatalf("invalid sequence number, got=%d", db.minSeq())
+	}
+	e2 = db.acquireSnapshot()
+	if db.minSeq() != 1 {
+		t.Fatalf("invalid sequence number, got=%d", db.minSeq())
+	}
+	db.releaseSnapshot(e1)
+	if db.minSeq() != 2 {
+		t.Fatalf("invalid sequence number, got=%d", db.minSeq())
+	}
+	db.releaseSnapshot(e2)
+	if db.minSeq() != 2 {
+		t.Fatalf("invalid sequence number, got=%d", db.minSeq())
+	}
+}
+
 func TestDb_HiddenValuesAreRemoved(t *testing.T) {
 	trun(t, func(h *dbHarness) {
 		s := h.db.s
 
 		h.put("foo", "v1")
 		h.compactMem()
-		m := kMaxMemCompactLevel
+		m := h.o.GetMaxMemCompationLevel()
 		v := s.version()
 		num := v.tLen(m)
 		v.release()
@@ -1177,7 +1236,7 @@ func TestDb_DeletionMarkers2(t *testing.T) {
 
 	h.put("foo", "v1")
 	h.compactMem()
-	m := kMaxMemCompactLevel
+	m := h.o.GetMaxMemCompationLevel()
 	v := s.version()
 	num := v.tLen(m)
 	v.release()
@@ -1212,7 +1271,7 @@ func TestDb_DeletionMarkers2(t *testing.T) {
 }
 
 func TestDb_CompactionTableOpenError(t *testing.T) {
-	h := newDbHarnessWopt(t, &opt.Options{MaxOpenFiles: 0})
+	h := newDbHarnessWopt(t, &opt.Options{CachedOpenFiles: -1})
 	defer h.close()
 
 	im := 10
@@ -1230,14 +1289,14 @@ func TestDb_CompactionTableOpenError(t *testing.T) {
 		t.Errorf("total tables is %d, want %d", n, im)
 	}
 
-	h.stor.SetOpenErr(storage.TypeTable)
+	h.stor.SetEmuErr(storage.TypeTable, tsOpOpen)
 	go h.db.CompactRange(util.Range{})
 	if err := h.db.compSendIdle(h.db.tcompCmdC); err != nil {
 		t.Log("compaction error: ", err)
 	}
 	h.closeDB0()
 	h.openDB()
-	h.stor.SetOpenErr(0)
+	h.stor.SetEmuErr(0, tsOpOpen)
 
 	for i := 0; i < im; i++ {
 		for j := 0; j < jm; j++ {
@@ -1248,7 +1307,7 @@ func TestDb_CompactionTableOpenError(t *testing.T) {
 
 func TestDb_OverlapInLevel0(t *testing.T) {
 	trun(t, func(h *dbHarness) {
-		if kMaxMemCompactLevel != 2 {
+		if h.o.GetMaxMemCompationLevel() != 2 {
 			t.Fatal("fix test to reflect the config")
 		}
 
@@ -1368,23 +1427,23 @@ func TestDb_ManifestWriteError(t *testing.T) {
 			h.compactMem()
 			h.getVal("foo", "bar")
 			v := h.db.s.version()
-			if n := v.tLen(kMaxMemCompactLevel); n != 1 {
+			if n := v.tLen(h.o.GetMaxMemCompationLevel()); n != 1 {
 				t.Errorf("invalid total tables, want=1 got=%d", n)
 			}
 			v.release()
 
 			if i == 0 {
-				h.stor.SetWriteErr(storage.TypeManifest)
+				h.stor.SetEmuErr(storage.TypeManifest, tsOpWrite)
 			} else {
-				h.stor.SetSyncErr(storage.TypeManifest)
+				h.stor.SetEmuErr(storage.TypeManifest, tsOpSync)
 			}
 
 			// Merging compaction (will fail)
-			h.compactRangeAtErr(kMaxMemCompactLevel, "", "", true)
+			h.compactRangeAtErr(h.o.GetMaxMemCompationLevel(), "", "", true)
 
 			h.db.Close()
-			h.stor.SetWriteErr(0)
-			h.stor.SetSyncErr(0)
+			h.stor.SetEmuErr(0, tsOpWrite)
+			h.stor.SetEmuErr(0, tsOpSync)
 
 			// Should not lose data
 			h.openDB()
@@ -1534,7 +1593,7 @@ func TestDb_ManualCompaction(t *testing.T) {
 	h := newDbHarness(t)
 	defer h.close()
 
-	if kMaxMemCompactLevel != 2 {
+	if h.o.GetMaxMemCompationLevel() != 2 {
 		t.Fatal("fix test to reflect the config")
 	}
 
@@ -1579,7 +1638,7 @@ func TestDb_BloomFilter(t *testing.T) {
 		return fmt.Sprintf("key%06d", i)
 	}
 
-	n := 10000
+	const n = 10000
 
 	// Populate multiple layers
 	for i := 0; i < n; i++ {
@@ -1818,7 +1877,7 @@ func TestDb_DeletionMarkersOnMemdb(t *testing.T) {
 }
 
 func TestDb_LeveldbIssue178(t *testing.T) {
-	nKeys := (kMaxTableSize / 30) * 5
+	nKeys := (opt.DefaultCompactionTableSize / 30) * 5
 	key1 := func(i int) string {
 		return fmt.Sprintf("my_key_%d", i)
 	}
@@ -1886,3 +1945,624 @@ func TestDb_LeveldbIssue200(t *testing.T) {
 	iter.Next()
 	assertBytes(t, []byte("5"), iter.Key())
 }
+
+func TestDb_GoleveldbIssue74(t *testing.T) {
+	h := newDbHarnessWopt(t, &opt.Options{
+		WriteBuffer: 1 * opt.MiB,
+	})
+	defer h.close()
+
+	const n, dur = 10000, 5 * time.Second
+
+	runtime.GOMAXPROCS(runtime.NumCPU())
+
+	until := time.Now().Add(dur)
+	wg := new(sync.WaitGroup)
+	wg.Add(2)
+	var done uint32
+	go func() {
+		var i int
+		defer func() {
+			t.Logf("WRITER DONE #%d", i)
+			atomic.StoreUint32(&done, 1)
+			wg.Done()
+		}()
+
+		b := new(Batch)
+		for ; time.Now().Before(until) && atomic.LoadUint32(&done) == 0; i++ {
+			iv := fmt.Sprintf("VAL%010d", i)
+			for k := 0; k < n; k++ {
+				key := fmt.Sprintf("KEY%06d", k)
+				b.Put([]byte(key), []byte(key+iv))
+				b.Put([]byte(fmt.Sprintf("PTR%06d", k)), []byte(key))
+			}
+			h.write(b)
+
+			b.Reset()
+			snap := h.getSnapshot()
+			iter := snap.NewIterator(util.BytesPrefix([]byte("PTR")), nil)
+			var k int
+			for ; iter.Next(); k++ {
+				ptrKey := iter.Key()
+				key := iter.Value()
+
+				if _, err := snap.Get(ptrKey, nil); err != nil {
+					t.Fatalf("WRITER #%d snapshot.Get %q: %v", i, ptrKey, err)
+				}
+				if value, err := snap.Get(key, nil); err != nil {
+					t.Fatalf("WRITER #%d snapshot.Get %q: %v", i, key, err)
+				} else if string(value) != string(key)+iv {
+					t.Fatalf("WRITER #%d snapshot.Get %q got invalid value, want %q got %q", i, key, string(key)+iv, value)
+				}
+
+				b.Delete(key)
+				b.Delete(ptrKey)
+			}
+			h.write(b)
+			iter.Release()
+			snap.Release()
+			if k != n {
+				t.Fatalf("#%d %d != %d", i, k, n)
+			}
+		}
+	}()
+	go func() {
+		var i int
+		defer func() {
+			t.Logf("READER DONE #%d", i)
+			atomic.StoreUint32(&done, 1)
+			wg.Done()
+		}()
+		for ; time.Now().Before(until) && atomic.LoadUint32(&done) == 0; i++ {
+			snap := h.getSnapshot()
+			iter := snap.NewIterator(util.BytesPrefix([]byte("PTR")), nil)
+			var prevValue string
+			var k int
+			for ; iter.Next(); k++ {
+				ptrKey := iter.Key()
+				key := iter.Value()
+
+				if _, err := snap.Get(ptrKey, nil); err != nil {
+					t.Fatalf("READER #%d snapshot.Get %q: %v", i, ptrKey, err)
+				}
+
+				if value, err := snap.Get(key, nil); err != nil {
+					t.Fatalf("READER #%d snapshot.Get %q: %v", i, key, err)
+				} else if prevValue != "" && string(value) != string(key)+prevValue {
+					t.Fatalf("READER #%d snapshot.Get %q got invalid value, want %q got %q", i, key, string(key)+prevValue, value)
+				} else {
+					prevValue = string(value[len(key):])
+				}
+			}
+			iter.Release()
+			snap.Release()
+			if k > 0 && k != n {
+				t.Fatalf("#%d %d != %d", i, k, n)
+			}
+		}
+	}()
+	wg.Wait()
+}
+
+func TestDb_GetProperties(t *testing.T) {
+	h := newDbHarness(t)
+	defer h.close()
+
+	_, err := h.db.GetProperty("leveldb.num-files-at-level")
+	if err == nil {
+		t.Error("GetProperty() failed to detect missing level")
+	}
+
+	_, err = h.db.GetProperty("leveldb.num-files-at-level0")
+	if err != nil {
+		t.Error("got unexpected error", err)
+	}
+
+	_, err = h.db.GetProperty("leveldb.num-files-at-level0x")
+	if err == nil {
+		t.Error("GetProperty() failed to detect invalid level")
+	}
+}
+
+func TestDb_GoleveldbIssue72and83(t *testing.T) {
+	h := newDbHarnessWopt(t, &opt.Options{
+		WriteBuffer:     1 * opt.MiB,
+		CachedOpenFiles: 3,
+	})
+	defer h.close()
+
+	const n, wn, dur = 10000, 100, 30 * time.Second
+
+	runtime.GOMAXPROCS(runtime.NumCPU())
+
+	randomData := func(prefix byte, i int) []byte {
+		data := make([]byte, 1+4+32+64+32)
+		_, err := crand.Reader.Read(data[1 : len(data)-4])
+		if err != nil {
+			panic(err)
+		}
+		data[0] = prefix
+		binary.LittleEndian.PutUint32(data[len(data)-4:], uint32(i))
+		return data
+	}
+
+	keys := make([][]byte, n)
+	for i := range keys {
+		keys[i] = randomData(1, 0)
+	}
+
+	until := time.Now().Add(dur)
+	wg := new(sync.WaitGroup)
+	wg.Add(3)
+	var done uint32
+	go func() {
+		i := 0
+		defer func() {
+			t.Logf("WRITER DONE #%d", i)
+			wg.Done()
+		}()
+
+		b := new(Batch)
+		for ; i < wn && atomic.LoadUint32(&done) == 0; i++ {
+			b.Reset()
+			for _, k1 := range keys {
+				k2 := randomData(2, i)
+				b.Put(k2, randomData(42, i))
+				b.Put(k1, k2)
+			}
+			if err := h.db.Write(b, h.wo); err != nil {
+				atomic.StoreUint32(&done, 1)
+				t.Fatalf("WRITER #%d db.Write: %v", i, err)
+			}
+		}
+	}()
+	go func() {
+		var i int
+		defer func() {
+			t.Logf("READER0 DONE #%d", i)
+			atomic.StoreUint32(&done, 1)
+			wg.Done()
+		}()
+		for ; time.Now().Before(until) && atomic.LoadUint32(&done) == 0; i++ {
+			snap := h.getSnapshot()
+			seq := snap.elem.seq
+			if seq == 0 {
+				snap.Release()
+				continue
+			}
+			iter := snap.NewIterator(util.BytesPrefix([]byte{1}), nil)
+			writei := int(snap.elem.seq/(n*2) - 1)
+			var k int
+			for ; iter.Next(); k++ {
+				k1 := iter.Key()
+				k2 := iter.Value()
+				kwritei := int(binary.LittleEndian.Uint32(k2[len(k2)-4:]))
+				if writei != kwritei {
+					t.Fatalf("READER0 #%d.%d W#%d invalid write iteration num: %d", i, k, writei, kwritei)
+				}
+				if _, err := snap.Get(k2, nil); err != nil {
+					t.Fatalf("READER0 #%d.%d W#%d snap.Get: %v\nk1: %x\n -> k2: %x", i, k, writei, err, k1, k2)
+				}
+			}
+			if err := iter.Error(); err != nil {
+				t.Fatalf("READER0 #%d.%d W#%d snap.Iterator: %v", i, k, writei, err)
+			}
+			iter.Release()
+			snap.Release()
+			if k > 0 && k != n {
+				t.Fatalf("READER0 #%d W#%d short read, got=%d want=%d", i, writei, k, n)
+			}
+		}
+	}()
+	go func() {
+		var i int
+		defer func() {
+			t.Logf("READER1 DONE #%d", i)
+			atomic.StoreUint32(&done, 1)
+			wg.Done()
+		}()
+		for ; time.Now().Before(until) && atomic.LoadUint32(&done) == 0; i++ {
+			iter := h.db.NewIterator(nil, nil)
+			seq := iter.(*dbIter).seq
+			if seq == 0 {
+				iter.Release()
+				continue
+			}
+			writei := int(seq/(n*2) - 1)
+			var k int
+			for ok := iter.Last(); ok; ok = iter.Prev() {
+				k++
+			}
+			if err := iter.Error(); err != nil {
+				t.Fatalf("READER1 #%d.%d W#%d db.Iterator: %v", i, k, writei, err)
+			}
+			iter.Release()
+			if m := (writei+1)*n + n; k != m {
+				t.Fatalf("READER1 #%d W#%d short read, got=%d want=%d", i, writei, k, m)
+			}
+		}
+	}()
+
+	wg.Wait()
+}
+
+func TestDb_TransientError(t *testing.T) {
+	h := newDbHarnessWopt(t, &opt.Options{
+		WriteBuffer:              128 * opt.KiB,
+		CachedOpenFiles:          3,
+		DisableCompactionBackoff: true,
+	})
+	defer h.close()
+
+	const (
+		nSnap = 20
+		nKey  = 10000
+	)
+
+	var (
+		snaps [nSnap]*Snapshot
+		b     = &Batch{}
+	)
+	for i := range snaps {
+		vtail := fmt.Sprintf("VAL%030d", i)
+		b.Reset()
+		for k := 0; k < nKey; k++ {
+			key := fmt.Sprintf("KEY%8d", k)
+			b.Put([]byte(key), []byte(key+vtail))
+		}
+		h.stor.SetEmuRandErr(storage.TypeTable, tsOpOpen, tsOpRead, tsOpReadAt)
+		if err := h.db.Write(b, nil); err != nil {
+			t.Logf("WRITE #%d error: %v", i, err)
+			h.stor.SetEmuRandErr(0, tsOpOpen, tsOpRead, tsOpReadAt, tsOpWrite)
+			for {
+				if err := h.db.Write(b, nil); err == nil {
+					break
+				} else if errors.IsCorrupted(err) {
+					t.Fatalf("WRITE #%d corrupted: %v", i, err)
+				}
+			}
+		}
+
+		snaps[i] = h.db.newSnapshot()
+		b.Reset()
+		for k := 0; k < nKey; k++ {
+			key := fmt.Sprintf("KEY%8d", k)
+			b.Delete([]byte(key))
+		}
+		h.stor.SetEmuRandErr(storage.TypeTable, tsOpOpen, tsOpRead, tsOpReadAt)
+		if err := h.db.Write(b, nil); err != nil {
+			t.Logf("WRITE #%d  error: %v", i, err)
+			h.stor.SetEmuRandErr(0, tsOpOpen, tsOpRead, tsOpReadAt)
+			for {
+				if err := h.db.Write(b, nil); err == nil {
+					break
+				} else if errors.IsCorrupted(err) {
+					t.Fatalf("WRITE #%d corrupted: %v", i, err)
+				}
+			}
+		}
+	}
+	h.stor.SetEmuRandErr(0, tsOpOpen, tsOpRead, tsOpReadAt)
+
+	runtime.GOMAXPROCS(runtime.NumCPU())
+
+	rnd := rand.New(rand.NewSource(0xecafdaed))
+	wg := &sync.WaitGroup{}
+	for i, snap := range snaps {
+		wg.Add(2)
+
+		go func(i int, snap *Snapshot, sk []int) {
+			defer wg.Done()
+
+			vtail := fmt.Sprintf("VAL%030d", i)
+			for _, k := range sk {
+				key := fmt.Sprintf("KEY%8d", k)
+				xvalue, err := snap.Get([]byte(key), nil)
+				if err != nil {
+					t.Fatalf("READER_GET #%d SEQ=%d K%d error: %v", i, snap.elem.seq, k, err)
+				}
+				value := key + vtail
+				if !bytes.Equal([]byte(value), xvalue) {
+					t.Fatalf("READER_GET #%d SEQ=%d K%d invalid value: want %q, got %q", i, snap.elem.seq, k, value, xvalue)
+				}
+			}
+		}(i, snap, rnd.Perm(nKey))
+
+		go func(i int, snap *Snapshot) {
+			defer wg.Done()
+
+			vtail := fmt.Sprintf("VAL%030d", i)
+			iter := snap.NewIterator(nil, nil)
+			defer iter.Release()
+			for k := 0; k < nKey; k++ {
+				if !iter.Next() {
+					if err := iter.Error(); err != nil {
+						t.Fatalf("READER_ITER #%d K%d error: %v", i, k, err)
+					} else {
+						t.Fatalf("READER_ITER #%d K%d eoi", i, k)
+					}
+				}
+				key := fmt.Sprintf("KEY%8d", k)
+				xkey := iter.Key()
+				if !bytes.Equal([]byte(key), xkey) {
+					t.Fatalf("READER_ITER #%d K%d invalid key: want %q, got %q", i, k, key, xkey)
+				}
+				value := key + vtail
+				xvalue := iter.Value()
+				if !bytes.Equal([]byte(value), xvalue) {
+					t.Fatalf("READER_ITER #%d K%d invalid value: want %q, got %q", i, k, value, xvalue)
+				}
+			}
+		}(i, snap)
+	}
+
+	wg.Wait()
+}
+
+func TestDb_UkeyShouldntHopAcrossTable(t *testing.T) {
+	h := newDbHarnessWopt(t, &opt.Options{
+		WriteBuffer:                 112 * opt.KiB,
+		CompactionTableSize:         90 * opt.KiB,
+		CompactionExpandLimitFactor: 1,
+	})
+	defer h.close()
+
+	const (
+		nSnap = 190
+		nKey  = 140
+	)
+
+	var (
+		snaps [nSnap]*Snapshot
+		b     = &Batch{}
+	)
+	for i := range snaps {
+		vtail := fmt.Sprintf("VAL%030d", i)
+		b.Reset()
+		for k := 0; k < nKey; k++ {
+			key := fmt.Sprintf("KEY%08d", k)
+			b.Put([]byte(key), []byte(key+vtail))
+		}
+		if err := h.db.Write(b, nil); err != nil {
+			t.Fatalf("WRITE #%d error: %v", i, err)
+		}
+
+		snaps[i] = h.db.newSnapshot()
+		b.Reset()
+		for k := 0; k < nKey; k++ {
+			key := fmt.Sprintf("KEY%08d", k)
+			b.Delete([]byte(key))
+		}
+		if err := h.db.Write(b, nil); err != nil {
+			t.Fatalf("WRITE #%d  error: %v", i, err)
+		}
+	}
+
+	h.compactMem()
+
+	h.waitCompaction()
+	for level, tables := range h.db.s.stVersion.tables {
+		for _, table := range tables {
+			t.Logf("L%d@%d %q:%q", level, table.file.Num(), table.imin, table.imax)
+		}
+	}
+
+	h.compactRangeAt(0, "", "")
+	h.waitCompaction()
+	for level, tables := range h.db.s.stVersion.tables {
+		for _, table := range tables {
+			t.Logf("L%d@%d %q:%q", level, table.file.Num(), table.imin, table.imax)
+		}
+	}
+	h.compactRangeAt(1, "", "")
+	h.waitCompaction()
+	for level, tables := range h.db.s.stVersion.tables {
+		for _, table := range tables {
+			t.Logf("L%d@%d %q:%q", level, table.file.Num(), table.imin, table.imax)
+		}
+	}
+	runtime.GOMAXPROCS(runtime.NumCPU())
+
+	wg := &sync.WaitGroup{}
+	for i, snap := range snaps {
+		wg.Add(1)
+
+		go func(i int, snap *Snapshot) {
+			defer wg.Done()
+
+			vtail := fmt.Sprintf("VAL%030d", i)
+			for k := 0; k < nKey; k++ {
+				key := fmt.Sprintf("KEY%08d", k)
+				xvalue, err := snap.Get([]byte(key), nil)
+				if err != nil {
+					t.Fatalf("READER_GET #%d SEQ=%d K%d error: %v", i, snap.elem.seq, k, err)
+				}
+				value := key + vtail
+				if !bytes.Equal([]byte(value), xvalue) {
+					t.Fatalf("READER_GET #%d SEQ=%d K%d invalid value: want %q, got %q", i, snap.elem.seq, k, value, xvalue)
+				}
+			}
+		}(i, snap)
+	}
+
+	wg.Wait()
+}
+
+func TestDb_TableCompactionBuilder(t *testing.T) {
+	stor := newTestStorage(t)
+	defer stor.Close()
+
+	const nSeq = 99
+
+	o := &opt.Options{
+		WriteBuffer:                 112 * opt.KiB,
+		CompactionTableSize:         43 * opt.KiB,
+		CompactionExpandLimitFactor: 1,
+		CompactionGPOverlapsFactor:  1,
+		BlockCache:                  opt.NoCache,
+	}
+	s, err := newSession(stor, o)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if err := s.create(); err != nil {
+		t.Fatal(err)
+	}
+	defer s.close()
+	var (
+		seq        uint64
+		targetSize = 5 * o.CompactionTableSize
+		value      = bytes.Repeat([]byte{'0'}, 100)
+	)
+	for i := 0; i < 2; i++ {
+		tw, err := s.tops.create()
+		if err != nil {
+			t.Fatal(err)
+		}
+		for k := 0; tw.tw.BytesLen() < targetSize; k++ {
+			key := []byte(fmt.Sprintf("%09d", k))
+			seq += nSeq - 1
+			for x := uint64(0); x < nSeq; x++ {
+				if err := tw.append(newIkey(key, seq-x, ktVal), value); err != nil {
+					t.Fatal(err)
+				}
+			}
+		}
+		tf, err := tw.finish()
+		if err != nil {
+			t.Fatal(err)
+		}
+		rec := &sessionRecord{numLevel: s.o.GetNumLevel()}
+		rec.addTableFile(i, tf)
+		if err := s.commit(rec); err != nil {
+			t.Fatal(err)
+		}
+	}
+
+	// Build grandparent.
+	v := s.version()
+	c := newCompaction(s, v, 1, append(tFiles{}, v.tables[1]...))
+	rec := &sessionRecord{numLevel: s.o.GetNumLevel()}
+	b := &tableCompactionBuilder{
+		s:         s,
+		c:         c,
+		rec:       rec,
+		stat1:     new(cStatsStaging),
+		minSeq:    0,
+		strict:    true,
+		tableSize: o.CompactionTableSize/3 + 961,
+	}
+	if err := b.run(new(compactionTransactCounter)); err != nil {
+		t.Fatal(err)
+	}
+	for _, t := range c.tables[0] {
+		rec.delTable(c.level, t.file.Num())
+	}
+	if err := s.commit(rec); err != nil {
+		t.Fatal(err)
+	}
+	c.release()
+
+	// Build level-1.
+	v = s.version()
+	c = newCompaction(s, v, 0, append(tFiles{}, v.tables[0]...))
+	rec = &sessionRecord{numLevel: s.o.GetNumLevel()}
+	b = &tableCompactionBuilder{
+		s:         s,
+		c:         c,
+		rec:       rec,
+		stat1:     new(cStatsStaging),
+		minSeq:    0,
+		strict:    true,
+		tableSize: o.CompactionTableSize,
+	}
+	if err := b.run(new(compactionTransactCounter)); err != nil {
+		t.Fatal(err)
+	}
+	for _, t := range c.tables[0] {
+		rec.delTable(c.level, t.file.Num())
+	}
+	// Move grandparent to level-3
+	for _, t := range v.tables[2] {
+		rec.delTable(2, t.file.Num())
+		rec.addTableFile(3, t)
+	}
+	if err := s.commit(rec); err != nil {
+		t.Fatal(err)
+	}
+	c.release()
+
+	v = s.version()
+	for level, want := range []bool{false, true, false, true, false} {
+		got := len(v.tables[level]) > 0
+		if want != got {
+			t.Fatalf("invalid level-%d tables len: want %v, got %v", level, want, got)
+		}
+	}
+	for i, f := range v.tables[1][:len(v.tables[1])-1] {
+		nf := v.tables[1][i+1]
+		if bytes.Equal(f.imax.ukey(), nf.imin.ukey()) {
+			t.Fatalf("KEY %q hop across table %d .. %d", f.imax.ukey(), f.file.Num(), nf.file.Num())
+		}
+	}
+	v.release()
+
+	// Compaction with transient error.
+	v = s.version()
+	c = newCompaction(s, v, 1, append(tFiles{}, v.tables[1]...))
+	rec = &sessionRecord{numLevel: s.o.GetNumLevel()}
+	b = &tableCompactionBuilder{
+		s:         s,
+		c:         c,
+		rec:       rec,
+		stat1:     new(cStatsStaging),
+		minSeq:    0,
+		strict:    true,
+		tableSize: o.CompactionTableSize,
+	}
+	stor.SetEmuErrOnce(storage.TypeTable, tsOpSync)
+	stor.SetEmuRandErr(storage.TypeTable, tsOpRead, tsOpReadAt, tsOpWrite)
+	stor.SetEmuRandErrProb(0xf0)
+	for {
+		if err := b.run(new(compactionTransactCounter)); err != nil {
+			t.Logf("(expected) b.run: %v", err)
+		} else {
+			break
+		}
+	}
+	if err := s.commit(rec); err != nil {
+		t.Fatal(err)
+	}
+	c.release()
+
+	stor.SetEmuErrOnce(0, tsOpSync)
+	stor.SetEmuRandErr(0, tsOpRead, tsOpReadAt, tsOpWrite)
+
+	v = s.version()
+	if len(v.tables[1]) != len(v.tables[2]) {
+		t.Fatalf("invalid tables length, want %d, got %d", len(v.tables[1]), len(v.tables[2]))
+	}
+	for i, f0 := range v.tables[1] {
+		f1 := v.tables[2][i]
+		iter0 := s.tops.newIterator(f0, nil, nil)
+		iter1 := s.tops.newIterator(f1, nil, nil)
+		for j := 0; true; j++ {
+			next0 := iter0.Next()
+			next1 := iter1.Next()
+			if next0 != next1 {
+				t.Fatalf("#%d.%d invalid eoi: want %v, got %v", i, j, next0, next1)
+			}
+			key0 := iter0.Key()
+			key1 := iter1.Key()
+			if !bytes.Equal(key0, key1) {
+				t.Fatalf("#%d.%d invalid key: want %q, got %q", i, j, key0, key1)
+			}
+			if next0 == false {
+				break
+			}
+		}
+		iter0.Release()
+		iter1.Release()
+	}
+	v.release()
+}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/db_util.go

@@ -7,8 +7,7 @@
 package leveldb
 
 import (
-	"errors"
-
+	"github.com/syndtr/goleveldb/leveldb/errors"
 	"github.com/syndtr/goleveldb/leveldb/iterator"
 	"github.com/syndtr/goleveldb/leveldb/opt"
 	"github.com/syndtr/goleveldb/leveldb/storage"
@@ -32,40 +31,44 @@ func (p Sizes) Sum() (n uint64) {
 	return n
 }
 
-// Check and clean files.
-func (d *DB) checkAndCleanFiles() error {
-	s := d.s
+// Logging.
+func (db *DB) log(v ...interface{})                 { db.s.log(v...) }
+func (db *DB) logf(format string, v ...interface{}) { db.s.logf(format, v...) }
 
-	v := s.version_NB()
-	tables := make(map[uint64]bool)
-	for _, tt := range v.tables {
-		for _, t := range tt {
-			tables[t.file.Num()] = false
+// Check and clean files.
+func (db *DB) checkAndCleanFiles() error {
+	v := db.s.version()
+	defer v.release()
+
+	tablesMap := make(map[uint64]bool)
+	for _, tables := range v.tables {
+		for _, t := range tables {
+			tablesMap[t.file.Num()] = false
 		}
 	}
 
-	ff, err := s.getFiles(storage.TypeAll)
+	files, err := db.s.getFiles(storage.TypeAll)
 	if err != nil {
 		return err
 	}
 
 	var nTables int
 	var rem []storage.File
-	for _, f := range ff {
+	for _, f := range files {
 		keep := true
 		switch f.Type() {
 		case storage.TypeManifest:
-			keep = f.Num() >= s.manifestFile.Num()
+			keep = f.Num() >= db.s.manifestFile.Num()
 		case storage.TypeJournal:
-			if d.frozenJournalFile != nil {
-				keep = f.Num() >= d.frozenJournalFile.Num()
+			if db.frozenJournalFile != nil {
+				keep = f.Num() >= db.frozenJournalFile.Num()
 			} else {
-				keep = f.Num() >= d.journalFile.Num()
+				keep = f.Num() >= db.journalFile.Num()
 			}
 		case storage.TypeTable:
-			_, keep = tables[f.Num()]
+			_, keep = tablesMap[f.Num()]
 			if keep {
-				tables[f.Num()] = true
+				tablesMap[f.Num()] = true
 				nTables++
 			}
 		}
@@ -75,18 +78,20 @@ func (d *DB) checkAndCleanFiles() error {
 		}
 	}
 
-	if nTables != len(tables) {
-		for num, present := range tables {
+	if nTables != len(tablesMap) {
+		var missing []*storage.FileInfo
+		for num, present := range tablesMap {
 			if !present {
-				s.logf("db@janitor table missing @%d", num)
+				missing = append(missing, &storage.FileInfo{Type: storage.TypeTable, Num: num})
+				db.logf("db@janitor table missing @%d", num)
 			}
 		}
-		return ErrCorrupted{Type: MissingFiles, Err: errors.New("leveldb: table files missing")}
+		return errors.NewErrCorrupted(nil, &errors.ErrMissingFiles{Files: missing})
 	}
 
-	s.logf("db@janitor F·%d G·%d", len(ff), len(rem))
+	db.logf("db@janitor F·%d G·%d", len(files), len(rem))
 	for _, f := range rem {
-		s.logf("db@janitor removing %s-%d", f.Type(), f.Num())
+		db.logf("db@janitor removing %s-%d", f.Type(), f.Num())
 		if err := f.Remove(); err != nil {
 			return err
 		}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/db_write.go

@@ -14,84 +14,93 @@ import (
 	"github.com/syndtr/goleveldb/leveldb/util"
 )
 
-func (d *DB) writeJournal(b *Batch) error {
-	w, err := d.journal.Next()
+func (db *DB) writeJournal(b *Batch) error {
+	w, err := db.journal.Next()
 	if err != nil {
 		return err
 	}
 	if _, err := w.Write(b.encode()); err != nil {
 		return err
 	}
-	if err := d.journal.Flush(); err != nil {
+	if err := db.journal.Flush(); err != nil {
 		return err
 	}
 	if b.sync {
-		return d.journalWriter.Sync()
+		return db.journalWriter.Sync()
 	}
 	return nil
 }
 
-func (d *DB) jWriter() {
-	defer d.closeW.Done()
+func (db *DB) jWriter() {
+	defer db.closeW.Done()
 	for {
 		select {
-		case b := <-d.journalC:
+		case b := <-db.journalC:
 			if b != nil {
-				d.journalAckC <- d.writeJournal(b)
+				db.journalAckC <- db.writeJournal(b)
 			}
-		case _, _ = <-d.closeC:
+		case _, _ = <-db.closeC:
 			return
 		}
 	}
 }
 
-func (d *DB) rotateMem(n int) (mem *memdb.DB, err error) {
+func (db *DB) rotateMem(n int) (mem *memDB, err error) {
 	// Wait for pending memdb compaction.
-	err = d.compSendIdle(d.mcompCmdC)
+	err = db.compSendIdle(db.mcompCmdC)
 	if err != nil {
 		return
 	}
 
 	// Create new memdb and journal.
-	mem, err = d.newMem(n)
+	mem, err = db.newMem(n)
 	if err != nil {
 		return
 	}
 
 	// Schedule memdb compaction.
-	d.compTrigger(d.mcompTriggerC)
+	db.compSendTrigger(db.mcompCmdC)
 	return
 }
 
-func (d *DB) flush(n int) (mem *memdb.DB, nn int, err error) {
-	s := d.s
-
+func (db *DB) flush(n int) (mem *memDB, nn int, err error) {
 	delayed := false
-	flush := func() bool {
-		v := s.version()
+	flush := func() (retry bool) {
+		v := db.s.version()
 		defer v.release()
-		mem = d.getEffectiveMem()
-		nn = mem.Free()
+		mem = db.getEffectiveMem()
+		defer func() {
+			if retry {
+				mem.decref()
+				mem = nil
+			}
+		}()
+		nn = mem.mdb.Free()
 		switch {
-		case v.tLen(0) >= kL0_SlowdownWritesTrigger && !delayed:
+		case v.tLen(0) >= db.s.o.GetWriteL0SlowdownTrigger() && !delayed:
 			delayed = true
 			time.Sleep(time.Millisecond)
 		case nn >= n:
 			return false
-		case v.tLen(0) >= kL0_StopWritesTrigger:
+		case v.tLen(0) >= db.s.o.GetWriteL0PauseTrigger():
 			delayed = true
-			err = d.compSendIdle(d.tcompCmdC)
+			err = db.compSendIdle(db.tcompCmdC)
 			if err != nil {
 				return false
 			}
 		default:
 			// Allow memdb to grow if it has no entry.
-			if mem.Len() == 0 {
+			if mem.mdb.Len() == 0 {
 				nn = n
-				return false
+			} else {
+				mem.decref()
+				mem, err = db.rotateMem(n)
+				if err == nil {
+					nn = mem.mdb.Free()
+				} else {
+					nn = 0
+				}
 			}
-			mem, err = d.rotateMem(n)
-			nn = mem.Free()
 			return false
 		}
 		return true
@@ -100,7 +109,12 @@ func (d *DB) flush(n int) (mem *memdb.DB, nn int, err error) {
 	for flush() {
 	}
 	if delayed {
-		s.logf("db@write delayed T·%v", time.Since(start))
+		db.writeDelay += time.Since(start)
+		db.writeDelayN++
+	} else if db.writeDelayN > 0 {
+		db.writeDelay = 0
+		db.writeDelayN = 0
+		db.logf("db@write was delayed N·%d T·%v", db.writeDelayN, db.writeDelay)
 	}
 	return
 }
@@ -109,39 +123,45 @@ func (d *DB) flush(n int) (mem *memdb.DB, nn int, err error) {
 // sequentially.
 //
 // It is safe to modify the contents of the arguments after Write returns.
-func (d *DB) Write(b *Batch, wo *opt.WriteOptions) (err error) {
-	err = d.ok()
-	if err != nil || b == nil || b.len() == 0 {
+func (db *DB) Write(b *Batch, wo *opt.WriteOptions) (err error) {
+	err = db.ok()
+	if err != nil || b == nil || b.Len() == 0 {
 		return
 	}
 
 	b.init(wo.GetSync())
 
 	// The write happen synchronously.
-retry:
 	select {
-	case d.writeC <- b:
-		if <-d.writeMergedC {
-			return <-d.writeAckC
+	case db.writeC <- b:
+		if <-db.writeMergedC {
+			return <-db.writeAckC
 		}
-		goto retry
-	case d.writeLockC <- struct{}{}:
-	case _, _ = <-d.closeC:
+	case db.writeLockC <- struct{}{}:
+	case err = <-db.compPerErrC:
+		return
+	case _, _ = <-db.closeC:
 		return ErrClosed
 	}
 
 	merged := 0
+	danglingMerge := false
 	defer func() {
-		<-d.writeLockC
+		if danglingMerge {
+			db.writeMergedC <- false
+		} else {
+			<-db.writeLockC
+		}
 		for i := 0; i < merged; i++ {
-			d.writeAckC <- err
+			db.writeAckC <- err
 		}
 	}()
 
-	mem, memFree, err := d.flush(b.size())
+	mem, memFree, err := db.flush(b.size())
 	if err != nil {
 		return
 	}
+	defer mem.decref()
 
 	// Calculate maximum size of the batch.
 	m := 1 << 20
@@ -154,13 +174,13 @@ retry:
 drain:
 	for b.size() < m && !b.sync {
 		select {
-		case nb := <-d.writeC:
+		case nb := <-db.writeC:
 			if b.size()+nb.size() <= m {
 				b.append(nb)
-				d.writeMergedC <- true
+				db.writeMergedC <- true
 				merged++
 			} else {
-				d.writeMergedC <- false
+				danglingMerge = true
 				break drain
 			}
 		default:
@@ -169,44 +189,52 @@ drain:
 	}
 
 	// Set batch first seq number relative from last seq.
-	b.seq = d.seq + 1
+	b.seq = db.seq + 1
 
 	// Write journal concurrently if it is large enough.
 	if b.size() >= (128 << 10) {
 		// Push the write batch to the journal writer
 		select {
-		case _, _ = <-d.closeC:
+		case db.journalC <- b:
+			// Write into memdb
+			if berr := b.memReplay(mem.mdb); berr != nil {
+				panic(berr)
+			}
+		case err = <-db.compPerErrC:
+			return
+		case _, _ = <-db.closeC:
 			err = ErrClosed
 			return
-		case d.journalC <- b:
-			// Write into memdb
-			b.memReplay(mem)
 		}
 		// Wait for journal writer
 		select {
-		case _, _ = <-d.closeC:
-			err = ErrClosed
-			return
-		case err = <-d.journalAckC:
+		case err = <-db.journalAckC:
 			if err != nil {
 				// Revert memdb if error detected
-				b.revertMemReplay(mem)
+				if berr := b.revertMemReplay(mem.mdb); berr != nil {
+					panic(berr)
+				}
 				return
 			}
+		case _, _ = <-db.closeC:
+			err = ErrClosed
+			return
 		}
 	} else {
-		err = d.writeJournal(b)
+		err = db.writeJournal(b)
 		if err != nil {
 			return
 		}
-		b.memReplay(mem)
+		if berr := b.memReplay(mem.mdb); berr != nil {
+			panic(berr)
+		}
 	}
 
 	// Set last seq number.
-	d.addSeq(uint64(b.len()))
+	db.addSeq(uint64(b.Len()))
 
 	if b.size() >= memFree {
-		d.rotateMem(0)
+		db.rotateMem(0)
 	}
 	return
 }
@@ -215,20 +243,20 @@ drain:
 // for that key; a DB is not a multi-map.
 //
 // It is safe to modify the contents of the arguments after Put returns.
-func (d *DB) Put(key, value []byte, wo *opt.WriteOptions) error {
+func (db *DB) Put(key, value []byte, wo *opt.WriteOptions) error {
 	b := new(Batch)
 	b.Put(key, value)
-	return d.Write(b, wo)
+	return db.Write(b, wo)
 }
 
 // Delete deletes the value for the given key. It returns ErrNotFound if
 // the DB does not contain the key.
 //
 // It is safe to modify the contents of the arguments after Delete returns.
-func (d *DB) Delete(key []byte, wo *opt.WriteOptions) error {
+func (db *DB) Delete(key []byte, wo *opt.WriteOptions) error {
 	b := new(Batch)
 	b.Delete(key)
-	return d.Write(b, wo)
+	return db.Write(b, wo)
 }
 
 func isMemOverlaps(icmp *iComparer, mem *memdb.DB, min, max []byte) bool {
@@ -247,33 +275,37 @@ func isMemOverlaps(icmp *iComparer, mem *memdb.DB, min, max []byte) bool {
 // A nil Range.Start is treated as a key before all keys in the DB.
 // And a nil Range.Limit is treated as a key after all keys in the DB.
 // Therefore if both is nil then it will compact entire DB.
-func (d *DB) CompactRange(r util.Range) error {
-	if err := d.ok(); err != nil {
+func (db *DB) CompactRange(r util.Range) error {
+	if err := db.ok(); err != nil {
 		return err
 	}
 
+	// Lock writer.
 	select {
-	case d.writeLockC <- struct{}{}:
-	case _, _ = <-d.closeC:
+	case db.writeLockC <- struct{}{}:
+	case err := <-db.compPerErrC:
+		return err
+	case _, _ = <-db.closeC:
 		return ErrClosed
 	}
 
 	// Check for overlaps in memdb.
-	mem := d.getEffectiveMem()
-	if isMemOverlaps(d.s.icmp, mem, r.Start, r.Limit) {
+	mem := db.getEffectiveMem()
+	defer mem.decref()
+	if isMemOverlaps(db.s.icmp, mem.mdb, r.Start, r.Limit) {
 		// Memdb compaction.
-		if _, err := d.rotateMem(0); err != nil {
-			<-d.writeLockC
+		if _, err := db.rotateMem(0); err != nil {
+			<-db.writeLockC
 			return err
 		}
-		<-d.writeLockC
-		if err := d.compSendIdle(d.mcompCmdC); err != nil {
+		<-db.writeLockC
+		if err := db.compSendIdle(db.mcompCmdC); err != nil {
 			return err
 		}
 	} else {
-		<-d.writeLockC
+		<-db.writeLockC
 	}
 
 	// Table compaction.
-	return d.compSendRange(d.tcompCmdC, -1, r.Start, r.Limit)
+	return db.compSendRange(db.tcompCmdC, -1, r.Start, r.Limit)
 }

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/doc.go

@@ -37,6 +37,16 @@
 //	err = iter.Error()
 //	...
 //
+// Iterate over subset of database content with a particular prefix:
+//	iter := db.NewIterator(util.BytesPrefix([]byte("foo-")), nil)
+//	for iter.Next() {
+//		// Use key/value.
+//		...
+//	}
+//	iter.Release()
+//	err = iter.Error()
+//	...
+//
 // Seek-then-Iterate:
 //
 // 	iter := db.NewIterator(nil, nil)

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/errors.go

@@ -7,32 +7,12 @@
 package leveldb
 
 import (
-	"errors"
-
-	"github.com/syndtr/goleveldb/leveldb/util"
+	"github.com/syndtr/goleveldb/leveldb/errors"
 )
 
 var (
-	ErrNotFound         = util.ErrNotFound
+	ErrNotFound         = errors.ErrNotFound
 	ErrSnapshotReleased = errors.New("leveldb: snapshot released")
 	ErrIterReleased     = errors.New("leveldb: iterator released")
 	ErrClosed           = errors.New("leveldb: closed")
 )
-
-type CorruptionType int
-
-const (
-	CorruptedManifest CorruptionType = iota
-	MissingFiles
-)
-
-// ErrCorrupted is the type that wraps errors that indicate corruption in
-// the database.
-type ErrCorrupted struct {
-	Type CorruptionType
-	Err  error
-}
-
-func (e ErrCorrupted) Error() string {
-	return e.Err.Error()
-}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/errors/errors.go

@@ -0,0 +1,76 @@
+// Copyright (c) 2014, Suryandaru Triandana <syndtr@gmail.com>
+// All rights reserved.
+//
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+// Package errors provides common error types used throughout leveldb.
+package errors
+
+import (
+	"errors"
+	"fmt"
+
+	"github.com/syndtr/goleveldb/leveldb/storage"
+	"github.com/syndtr/goleveldb/leveldb/util"
+)
+
+var (
+	ErrNotFound    = New("leveldb: not found")
+	ErrReleased    = util.ErrReleased
+	ErrHasReleaser = util.ErrHasReleaser
+)
+
+// New returns an error that formats as the given text.
+func New(text string) error {
+	return errors.New(text)
+}
+
+// ErrCorrupted is the type that wraps errors that indicate corruption in
+// the database.
+type ErrCorrupted struct {
+	File *storage.FileInfo
+	Err  error
+}
+
+func (e *ErrCorrupted) Error() string {
+	if e.File != nil {
+		return fmt.Sprintf("%v [file=%v]", e.Err, e.File)
+	} else {
+		return e.Err.Error()
+	}
+}
+
+// NewErrCorrupted creates new ErrCorrupted error.
+func NewErrCorrupted(f storage.File, err error) error {
+	return &ErrCorrupted{storage.NewFileInfo(f), err}
+}
+
+// IsCorrupted returns a boolean indicating whether the error is indicating
+// a corruption.
+func IsCorrupted(err error) bool {
+	switch err.(type) {
+	case *ErrCorrupted:
+		return true
+	}
+	return false
+}
+
+// ErrMissingFiles is the type that indicating a corruption due to missing
+// files.
+type ErrMissingFiles struct {
+	Files []*storage.FileInfo
+}
+
+func (e *ErrMissingFiles) Error() string { return "file missing" }
+
+// SetFile sets 'file info' of the given error with the given file.
+// Currently only ErrCorrupted is supported, otherwise will do nothing.
+func SetFile(err error, f storage.File) error {
+	switch x := err.(type) {
+	case *ErrCorrupted:
+		x.File = storage.NewFileInfo(f)
+		return x
+	}
+	return err
+}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/external_test.go

@@ -21,7 +21,7 @@ var _ = testutil.Defer(func() {
 			BlockRestartInterval: 5,
 			BlockSize:            50,
 			Compression:          opt.NoCompression,
-			MaxOpenFiles:         0,
+			CachedOpenFiles:      -1,
 			Strict:               opt.StrictAll,
 			WriteBuffer:          1000,
 		}
@@ -36,22 +36,21 @@ var _ = testutil.Defer(func() {
 				testutil.DoDBTesting(&t)
 				db.TestClose()
 				done <- true
-			}, 9.0)
+			}, 20.0)
 		})
 
 		Describe("read test", func() {
-			testutil.AllKeyValueTesting(nil, func(kv testutil.KeyValue) testutil.DB {
+			testutil.AllKeyValueTesting(nil, nil, func(kv testutil.KeyValue) testutil.DB {
 				// Building the DB.
 				db := newTestingDB(o, nil, nil)
 				kv.IterateShuffled(nil, func(i int, key, value []byte) {
 					err := db.TestPut(key, value)
 					Expect(err).NotTo(HaveOccurred())
 				})
-				testutil.Defer("teardown", func() {
-					db.TestClose()
-				})
 
 				return db
+			}, func(db testutil.DB) {
+				db.(*testingDB).TestClose()
 			})
 		})
 	})

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/go13_bench_test.go

@@ -0,0 +1,58 @@
+// Copyright (c) 2012, Suryandaru Triandana <syndtr@gmail.com>
+// All rights reserved.
+//
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+// +build go1.3
+
+package leveldb
+
+import (
+	"sync/atomic"
+	"testing"
+)
+
+func BenchmarkDBReadConcurrent(b *testing.B) {
+	p := openDBBench(b, false)
+	p.populate(b.N)
+	p.fill()
+	p.gc()
+	defer p.close()
+
+	b.ResetTimer()
+	b.SetBytes(116)
+
+	b.RunParallel(func(pb *testing.PB) {
+		iter := p.newIter()
+		defer iter.Release()
+		for pb.Next() && iter.Next() {
+		}
+	})
+}
+
+func BenchmarkDBReadConcurrent2(b *testing.B) {
+	p := openDBBench(b, false)
+	p.populate(b.N)
+	p.fill()
+	p.gc()
+	defer p.close()
+
+	b.ResetTimer()
+	b.SetBytes(116)
+
+	var dir uint32
+	b.RunParallel(func(pb *testing.PB) {
+		iter := p.newIter()
+		defer iter.Release()
+		if atomic.AddUint32(&dir, 1)%2 == 0 {
+			for pb.Next() && iter.Next() {
+			}
+		} else {
+			if pb.Next() && iter.Last() {
+				for pb.Next() && iter.Prev() {
+				}
+			}
+		}
+	})
+}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/iterator/array_iter.go

@@ -40,13 +40,19 @@ type basicArrayIterator struct {
 	util.BasicReleaser
 	array BasicArray
 	pos   int
+	err   error
 }
 
 func (i *basicArrayIterator) Valid() bool {
-	return i.pos >= 0 && i.pos < i.array.Len()
+	return i.pos >= 0 && i.pos < i.array.Len() && !i.Released()
 }
 
 func (i *basicArrayIterator) First() bool {
+	if i.Released() {
+		i.err = ErrIterReleased
+		return false
+	}
+
 	if i.array.Len() == 0 {
 		i.pos = -1
 		return false
@@ -56,6 +62,11 @@ func (i *basicArrayIterator) First() bool {
 }
 
 func (i *basicArrayIterator) Last() bool {
+	if i.Released() {
+		i.err = ErrIterReleased
+		return false
+	}
+
 	n := i.array.Len()
 	if n == 0 {
 		i.pos = 0
@@ -66,6 +77,11 @@ func (i *basicArrayIterator) Last() bool {
 }
 
 func (i *basicArrayIterator) Seek(key []byte) bool {
+	if i.Released() {
+		i.err = ErrIterReleased
+		return false
+	}
+
 	n := i.array.Len()
 	if n == 0 {
 		i.pos = 0
@@ -79,6 +95,11 @@ func (i *basicArrayIterator) Seek(key []byte) bool {
 }
 
 func (i *basicArrayIterator) Next() bool {
+	if i.Released() {
+		i.err = ErrIterReleased
+		return false
+	}
+
 	i.pos++
 	if n := i.array.Len(); i.pos >= n {
 		i.pos = n
@@ -88,6 +109,11 @@ func (i *basicArrayIterator) Next() bool {
 }
 
 func (i *basicArrayIterator) Prev() bool {
+	if i.Released() {
+		i.err = ErrIterReleased
+		return false
+	}
+
 	i.pos--
 	if i.pos < 0 {
 		i.pos = -1
@@ -96,7 +122,7 @@ func (i *basicArrayIterator) Prev() bool {
 	return true
 }
 
-func (i *basicArrayIterator) Error() error { return nil }
+func (i *basicArrayIterator) Error() error { return i.err }
 
 type arrayIterator struct {
 	basicArrayIterator

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/iterator/indexed_iter.go

@@ -7,6 +7,7 @@
 package iterator
 
 import (
+	"github.com/syndtr/goleveldb/leveldb/errors"
 	"github.com/syndtr/goleveldb/leveldb/util"
 )
 
@@ -22,13 +23,13 @@ type IteratorIndexer interface {
 
 type indexedIterator struct {
 	util.BasicReleaser
-	index     IteratorIndexer
-	strict    bool
-	strictGet bool
+	index  IteratorIndexer
+	strict bool
 
-	data Iterator
-	err  error
-	errf func(err error)
+	data   Iterator
+	err    error
+	errf   func(err error)
+	closed bool
 }
 
 func (i *indexedIterator) setData() {
@@ -36,11 +37,6 @@ func (i *indexedIterator) setData() {
 		i.data.Release()
 	}
 	i.data = i.index.Get()
-	if i.strictGet {
-		if err := i.data.Error(); err != nil {
-			i.err = err
-		}
-	}
 }
 
 func (i *indexedIterator) clearData() {
@@ -50,14 +46,21 @@ func (i *indexedIterator) clearData() {
 	i.data = nil
 }
 
-func (i *indexedIterator) dataErr() bool {
-	if i.errf != nil {
-		if err := i.data.Error(); err != nil {
+func (i *indexedIterator) indexErr() {
+	if err := i.index.Error(); err != nil {
+		if i.errf != nil {
 			i.errf(err)
 		}
+		i.err = err
 	}
-	if i.strict {
-		if err := i.data.Error(); err != nil {
+}
+
+func (i *indexedIterator) dataErr() bool {
+	if err := i.data.Error(); err != nil {
+		if i.errf != nil {
+			i.errf(err)
+		}
+		if i.strict || !errors.IsCorrupted(err) {
 			i.err = err
 			return true
 		}
@@ -72,9 +75,13 @@ func (i *indexedIterator) Valid() bool {
 func (i *indexedIterator) First() bool {
 	if i.err != nil {
 		return false
+	} else if i.Released() {
+		i.err = ErrIterReleased
+		return false
 	}
 
 	if !i.index.First() {
+		i.indexErr()
 		i.clearData()
 		return false
 	}
@@ -85,9 +92,13 @@ func (i *indexedIterator) First() bool {
 func (i *indexedIterator) Last() bool {
 	if i.err != nil {
 		return false
+	} else if i.Released() {
+		i.err = ErrIterReleased
+		return false
 	}
 
 	if !i.index.Last() {
+		i.indexErr()
 		i.clearData()
 		return false
 	}
@@ -105,9 +116,13 @@ func (i *indexedIterator) Last() bool {
 func (i *indexedIterator) Seek(key []byte) bool {
 	if i.err != nil {
 		return false
+	} else if i.Released() {
+		i.err = ErrIterReleased
+		return false
 	}
 
 	if !i.index.Seek(key) {
+		i.indexErr()
 		i.clearData()
 		return false
 	}
@@ -125,6 +140,9 @@ func (i *indexedIterator) Seek(key []byte) bool {
 func (i *indexedIterator) Next() bool {
 	if i.err != nil {
 		return false
+	} else if i.Released() {
+		i.err = ErrIterReleased
+		return false
 	}
 
 	switch {
@@ -136,6 +154,7 @@ func (i *indexedIterator) Next() bool {
 		fallthrough
 	case i.data == nil:
 		if !i.index.Next() {
+			i.indexErr()
 			return false
 		}
 		i.setData()
@@ -147,6 +166,9 @@ func (i *indexedIterator) Next() bool {
 func (i *indexedIterator) Prev() bool {
 	if i.err != nil {
 		return false
+	} else if i.Released() {
+		i.err = ErrIterReleased
+		return false
 	}
 
 	switch {
@@ -158,6 +180,7 @@ func (i *indexedIterator) Prev() bool {
 		fallthrough
 	case i.data == nil:
 		if !i.index.Prev() {
+			i.indexErr()
 			return false
 		}
 		i.setData()
@@ -206,16 +229,14 @@ func (i *indexedIterator) SetErrorCallback(f func(err error)) {
 	i.errf = f
 }
 
-// NewIndexedIterator returns an indexed iterator. An index is iterator
-// that returns another iterator, a data iterator. A data iterator is the
+// NewIndexedIterator returns an 'indexed iterator'. An index is iterator
+// that returns another iterator, a 'data iterator'. A 'data iterator' is the
 // iterator that contains actual key/value pairs.
 //
-// If strict is true then error yield by data iterator will halt the indexed
-// iterator, on contrary if strict is false then the indexed iterator will
-// ignore those error and move on to the next index. If strictGet is true and
-// index.Get() yield an 'error iterator' then the indexed iterator will be halted.
-// An 'error iterator' is iterator which its Error() method always return non-nil
-// even before any 'seeks method' is called.
-func NewIndexedIterator(index IteratorIndexer, strict, strictGet bool) Iterator {
-	return &indexedIterator{index: index, strict: strict, strictGet: strictGet}
+// If strict is true the any 'corruption errors' (i.e errors.IsCorrupted(err) == true)
+// won't be ignored and will halt 'indexed iterator', otherwise the iterator will
+// continue to the next 'data iterator'. Corruption on 'index iterator' will not be
+// ignored and will halt the iterator.
+func NewIndexedIterator(index IteratorIndexer, strict bool) Iterator {
+	return &indexedIterator{index: index, strict: strict}
 }

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/iterator/indexed_iter_test.go

@@ -65,7 +65,7 @@ var _ = testutil.Defer(func() {
 					// Test the iterator.
 					t := testutil.IteratorTesting{
 						KeyValue: kv.Clone(),
-						Iter:     NewIndexedIterator(NewArrayIndexer(index), true, true),
+						Iter:     NewIndexedIterator(NewArrayIndexer(index), true),
 					}
 					testutil.DoIteratorTesting(&t)
 					done <- true

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/iterator/iter.go

@@ -14,6 +14,10 @@ import (
 	"github.com/syndtr/goleveldb/leveldb/util"
 )
 
+var (
+	ErrIterReleased = errors.New("leveldb/iterator: iterator released")
+)
+
 // IteratorSeeker is the interface that wraps the 'seeks method'.
 type IteratorSeeker interface {
 	// First moves the iterator to the first key/value pair. If the iterator
@@ -100,28 +104,13 @@ type ErrorCallbackSetter interface {
 }
 
 type emptyIterator struct {
-	releaser util.Releaser
-	released bool
-	err      error
+	util.BasicReleaser
+	err error
 }
 
 func (i *emptyIterator) rErr() {
-	if i.err == nil && i.released {
-		i.err = errors.New("leveldb/iterator: iterator released")
-	}
-}
-
-func (i *emptyIterator) Release() {
-	if i.releaser != nil {
-		i.releaser.Release()
-		i.releaser = nil
-	}
-	i.released = true
-}
-
-func (i *emptyIterator) SetReleaser(releaser util.Releaser) {
-	if !i.released {
-		i.releaser = releaser
+	if i.err == nil && i.Released() {
+		i.err = ErrIterReleased
 	}
 }
 

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/iterator/merged_iter.go

@@ -7,16 +7,11 @@
 package iterator
 
 import (
-	"errors"
-
 	"github.com/syndtr/goleveldb/leveldb/comparer"
+	"github.com/syndtr/goleveldb/leveldb/errors"
 	"github.com/syndtr/goleveldb/leveldb/util"
 )
 
-var (
-	ErrIterReleased = errors.New("leveldb/iterator: iterator released")
-)
-
 type dir int
 
 const (
@@ -48,13 +43,11 @@ func assertKey(key []byte) []byte {
 }
 
 func (i *mergedIterator) iterErr(iter Iterator) bool {
-	if i.errf != nil {
-		if err := iter.Error(); err != nil {
+	if err := iter.Error(); err != nil {
+		if i.errf != nil {
 			i.errf(err)
 		}
-	}
-	if i.strict {
-		if err := iter.Error(); err != nil {
+		if i.strict || !errors.IsCorrupted(err) {
 			i.err = err
 			return true
 		}
@@ -274,9 +267,13 @@ func (i *mergedIterator) Release() {
 }
 
 func (i *mergedIterator) SetReleaser(releaser util.Releaser) {
-	if i.dir != dirReleased {
-		i.releaser = releaser
+	if i.dir == dirReleased {
+		panic(util.ErrReleased)
 	}
+	if i.releaser != nil && releaser != nil {
+		panic(util.ErrHasReleaser)
+	}
+	i.releaser = releaser
 }
 
 func (i *mergedIterator) Error() error {
@@ -294,9 +291,9 @@ func (i *mergedIterator) SetErrorCallback(f func(err error)) {
 // keys: if iters[i] contains a key k then iters[j] will not contain that key k.
 // None of the iters may be nil.
 //
-// If strict is true then error yield by any iterators will halt the merged
-// iterator, on contrary if strict is false then the merged iterator will
-// ignore those error and move on to the next iterator.
+// If strict is true the any 'corruption errors' (i.e errors.IsCorrupted(err) == true)
+// won't be ignored and will halt 'merged iterator', otherwise the iterator will
+// continue to the next 'input iterator'.
 func NewMergedIterator(iters []Iterator, cmp comparer.Comparer, strict bool) Iterator {
 	return &mergedIterator{
 		iters:  iters,

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/journal/journal.go

@@ -79,10 +79,10 @@ package journal
 
 import (
 	"encoding/binary"
-	"errors"
 	"fmt"
 	"io"
 
+	"github.com/syndtr/goleveldb/leveldb/errors"
 	"github.com/syndtr/goleveldb/leveldb/util"
 )
 
@@ -103,18 +103,18 @@ type flusher interface {
 	Flush() error
 }
 
-// DroppedError is the error type that passed to Dropper.Drop method.
-type DroppedError struct {
+// ErrCorrupted is the error type that generated by corrupted block or chunk.
+type ErrCorrupted struct {
 	Size   int
 	Reason string
 }
 
-func (e DroppedError) Error() string {
-	return fmt.Sprintf("leveldb/journal: dropped %d bytes: %s", e.Size, e.Reason)
+func (e *ErrCorrupted) Error() string {
+	return fmt.Sprintf("leveldb/journal: block/chunk corrupted: %s (%d bytes)", e.Reason, e.Size)
 }
 
 // Dropper is the interface that wrap simple Drop method. The Drop
-// method will be called when the journal reader dropping a chunk.
+// method will be called when the journal reader dropping a block or chunk.
 type Dropper interface {
 	Drop(err error)
 }
@@ -158,76 +158,78 @@ func NewReader(r io.Reader, dropper Dropper, strict, checksum bool) *Reader {
 	}
 }
 
+var errSkip = errors.New("leveldb/journal: skipped")
+
+func (r *Reader) corrupt(n int, reason string, skip bool) error {
+	if r.dropper != nil {
+		r.dropper.Drop(&ErrCorrupted{n, reason})
+	}
+	if r.strict && !skip {
+		r.err = errors.NewErrCorrupted(nil, &ErrCorrupted{n, reason})
+		return r.err
+	}
+	return errSkip
+}
+
 // nextChunk sets r.buf[r.i:r.j] to hold the next chunk's payload, reading the
 // next block into the buffer if necessary.
-func (r *Reader) nextChunk(wantFirst, skip bool) error {
+func (r *Reader) nextChunk(first bool) error {
 	for {
 		if r.j+headerSize <= r.n {
 			checksum := binary.LittleEndian.Uint32(r.buf[r.j+0 : r.j+4])
 			length := binary.LittleEndian.Uint16(r.buf[r.j+4 : r.j+6])
 			chunkType := r.buf[r.j+6]
 
-			var err error
 			if checksum == 0 && length == 0 && chunkType == 0 {
 				// Drop entire block.
-				err = DroppedError{r.n - r.j, "zero header"}
+				m := r.n - r.j
 				r.i = r.n
 				r.j = r.n
+				return r.corrupt(m, "zero header", false)
 			} else {
 				m := r.n - r.j
 				r.i = r.j + headerSize
 				r.j = r.j + headerSize + int(length)
 				if r.j > r.n {
 					// Drop entire block.
-					err = DroppedError{m, "chunk length overflows block"}
 					r.i = r.n
 					r.j = r.n
+					return r.corrupt(m, "chunk length overflows block", false)
 				} else if r.checksum && checksum != util.NewCRC(r.buf[r.i-1:r.j]).Value() {
 					// Drop entire block.
-					err = DroppedError{m, "checksum mismatch"}
 					r.i = r.n
 					r.j = r.n
+					return r.corrupt(m, "checksum mismatch", false)
 				}
 			}
-			if wantFirst && err == nil && chunkType != fullChunkType && chunkType != firstChunkType {
-				if skip {
-					// The chunk are intentionally skipped.
-					if chunkType == lastChunkType {
-						skip = false
-					}
-					continue
-				} else {
-					// Drop the chunk.
-					err = DroppedError{r.j - r.i + headerSize, "orphan chunk"}
-				}
+			if first && chunkType != fullChunkType && chunkType != firstChunkType {
+				m := r.j - r.i
+				r.i = r.j
+				// Report the error, but skip it.
+				return r.corrupt(m+headerSize, "orphan chunk", true)
 			}
-			if err == nil {
-				r.last = chunkType == fullChunkType || chunkType == lastChunkType
-			} else {
-				if r.dropper != nil {
-					r.dropper.Drop(err)
-				}
-				if r.strict {
-					r.err = err
-				}
+			r.last = chunkType == fullChunkType || chunkType == lastChunkType
+			return nil
+		}
+
+		// The last block.
+		if r.n < blockSize && r.n > 0 {
+			if !first {
+				return r.corrupt(0, "missing chunk part", false)
 			}
+			r.err = io.EOF
+			return r.err
+		}
+
+		// Read block.
+		n, err := io.ReadFull(r.r, r.buf[:])
+		if err != nil && err != io.EOF && err != io.ErrUnexpectedEOF {
 			return err
 		}
-		if r.n < blockSize && r.n > 0 {
-			// This is the last block.
-			if r.j != r.n {
-				r.err = io.ErrUnexpectedEOF
-			} else {
-				r.err = io.EOF
-			}
-			return r.err
-		}
-		n, err := io.ReadFull(r.r, r.buf[:])
-		if err != nil && err != io.ErrUnexpectedEOF {
-			r.err = err
-			return r.err
-		}
 		if n == 0 {
+			if !first {
+				return r.corrupt(0, "missing chunk part", false)
+			}
 			r.err = io.EOF
 			return r.err
 		}
@@ -237,29 +239,26 @@ func (r *Reader) nextChunk(wantFirst, skip bool) error {
 
 // Next returns a reader for the next journal. It returns io.EOF if there are no
 // more journals. The reader returned becomes stale after the next Next call,
-// and should no longer be used.
+// and should no longer be used. If strict is false, the reader will returns
+// io.ErrUnexpectedEOF error when found corrupted journal.
 func (r *Reader) Next() (io.Reader, error) {
 	r.seq++
 	if r.err != nil {
 		return nil, r.err
 	}
-	skip := !r.last
+	r.i = r.j
 	for {
-		r.i = r.j
-		if r.nextChunk(true, skip) != nil {
-			// So that 'orphan chunk' drop will be reported.
-			skip = false
-		} else {
+		if err := r.nextChunk(true); err == nil {
 			break
-		}
-		if r.err != nil {
-			return nil, r.err
+		} else if err != errSkip {
+			return nil, err
 		}
 	}
 	return &singleReader{r, r.seq, nil}, nil
 }
 
-// Reset resets the journal reader, allows reuse of the journal reader.
+// Reset resets the journal reader, allows reuse of the journal reader. Reset returns
+// last accumulated error.
 func (r *Reader) Reset(reader io.Reader, dropper Dropper, strict, checksum bool) error {
 	r.seq++
 	err := r.err
@@ -296,7 +295,11 @@ func (x *singleReader) Read(p []byte) (int, error) {
 		if r.last {
 			return 0, io.EOF
 		}
-		if x.err = r.nextChunk(false, false); x.err != nil {
+		x.err = r.nextChunk(false)
+		if x.err != nil {
+			if x.err == errSkip {
+				x.err = io.ErrUnexpectedEOF
+			}
 			return 0, x.err
 		}
 	}
@@ -320,7 +323,11 @@ func (x *singleReader) ReadByte() (byte, error) {
 		if r.last {
 			return 0, io.EOF
 		}
-		if x.err = r.nextChunk(false, false); x.err != nil {
+		x.err = r.nextChunk(false)
+		if x.err != nil {
+			if x.err == errSkip {
+				x.err = io.ErrUnexpectedEOF
+			}
 			return 0, x.err
 		}
 	}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/journal/journal_test.go

@@ -12,6 +12,7 @@ package journal
 
 import (
 	"bytes"
+	"encoding/binary"
 	"fmt"
 	"io"
 	"io/ioutil"
@@ -326,3 +327,492 @@ func TestStaleWriter(t *testing.T) {
 		t.Fatalf("stale write #1: unexpected error: %v", err)
 	}
 }
+
+func TestCorrupt_MissingLastBlock(t *testing.T) {
+	buf := new(bytes.Buffer)
+
+	w := NewWriter(buf)
+
+	// First record.
+	ww, err := w.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := ww.Write(bytes.Repeat([]byte("0"), blockSize-1024)); err != nil {
+		t.Fatalf("write #0: unexpected error: %v", err)
+	}
+
+	// Second record.
+	ww, err = w.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := ww.Write(bytes.Repeat([]byte("0"), blockSize-headerSize)); err != nil {
+		t.Fatalf("write #1: unexpected error: %v", err)
+	}
+
+	if err := w.Close(); err != nil {
+		t.Fatal(err)
+	}
+
+	// Cut the last block.
+	b := buf.Bytes()[:blockSize]
+	r := NewReader(bytes.NewReader(b), dropper{t}, false, true)
+
+	// First read.
+	rr, err := r.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	n, err := io.Copy(ioutil.Discard, rr)
+	if err != nil {
+		t.Fatalf("read #0: %v", err)
+	}
+	if n != blockSize-1024 {
+		t.Fatalf("read #0: got %d bytes want %d", n, blockSize-1024)
+	}
+
+	// Second read.
+	rr, err = r.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	n, err = io.Copy(ioutil.Discard, rr)
+	if err != io.ErrUnexpectedEOF {
+		t.Fatalf("read #1: unexpected error: %v", err)
+	}
+
+	if _, err := r.Next(); err != io.EOF {
+		t.Fatalf("last next: unexpected error: %v", err)
+	}
+}
+
+func TestCorrupt_CorruptedFirstBlock(t *testing.T) {
+	buf := new(bytes.Buffer)
+
+	w := NewWriter(buf)
+
+	// First record.
+	ww, err := w.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := ww.Write(bytes.Repeat([]byte("0"), blockSize/2)); err != nil {
+		t.Fatalf("write #0: unexpected error: %v", err)
+	}
+
+	// Second record.
+	ww, err = w.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := ww.Write(bytes.Repeat([]byte("0"), blockSize-headerSize)); err != nil {
+		t.Fatalf("write #1: unexpected error: %v", err)
+	}
+
+	// Third record.
+	ww, err = w.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := ww.Write(bytes.Repeat([]byte("0"), (blockSize-headerSize)+1)); err != nil {
+		t.Fatalf("write #2: unexpected error: %v", err)
+	}
+
+	// Fourth record.
+	ww, err = w.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := ww.Write(bytes.Repeat([]byte("0"), (blockSize-headerSize)+2)); err != nil {
+		t.Fatalf("write #3: unexpected error: %v", err)
+	}
+
+	if err := w.Close(); err != nil {
+		t.Fatal(err)
+	}
+
+	b := buf.Bytes()
+	// Corrupting block #0.
+	for i := 0; i < 1024; i++ {
+		b[i] = '1'
+	}
+
+	r := NewReader(bytes.NewReader(b), dropper{t}, false, true)
+
+	// First read (third record).
+	rr, err := r.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	n, err := io.Copy(ioutil.Discard, rr)
+	if err != nil {
+		t.Fatalf("read #0: %v", err)
+	}
+	if want := int64(blockSize-headerSize) + 1; n != want {
+		t.Fatalf("read #0: got %d bytes want %d", n, want)
+	}
+
+	// Second read (fourth record).
+	rr, err = r.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	n, err = io.Copy(ioutil.Discard, rr)
+	if err != nil {
+		t.Fatalf("read #1: %v", err)
+	}
+	if want := int64(blockSize-headerSize) + 2; n != want {
+		t.Fatalf("read #1: got %d bytes want %d", n, want)
+	}
+
+	if _, err := r.Next(); err != io.EOF {
+		t.Fatalf("last next: unexpected error: %v", err)
+	}
+}
+
+func TestCorrupt_CorruptedMiddleBlock(t *testing.T) {
+	buf := new(bytes.Buffer)
+
+	w := NewWriter(buf)
+
+	// First record.
+	ww, err := w.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := ww.Write(bytes.Repeat([]byte("0"), blockSize/2)); err != nil {
+		t.Fatalf("write #0: unexpected error: %v", err)
+	}
+
+	// Second record.
+	ww, err = w.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := ww.Write(bytes.Repeat([]byte("0"), blockSize-headerSize)); err != nil {
+		t.Fatalf("write #1: unexpected error: %v", err)
+	}
+
+	// Third record.
+	ww, err = w.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := ww.Write(bytes.Repeat([]byte("0"), (blockSize-headerSize)+1)); err != nil {
+		t.Fatalf("write #2: unexpected error: %v", err)
+	}
+
+	// Fourth record.
+	ww, err = w.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := ww.Write(bytes.Repeat([]byte("0"), (blockSize-headerSize)+2)); err != nil {
+		t.Fatalf("write #3: unexpected error: %v", err)
+	}
+
+	if err := w.Close(); err != nil {
+		t.Fatal(err)
+	}
+
+	b := buf.Bytes()
+	// Corrupting block #1.
+	for i := 0; i < 1024; i++ {
+		b[blockSize+i] = '1'
+	}
+
+	r := NewReader(bytes.NewReader(b), dropper{t}, false, true)
+
+	// First read (first record).
+	rr, err := r.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	n, err := io.Copy(ioutil.Discard, rr)
+	if err != nil {
+		t.Fatalf("read #0: %v", err)
+	}
+	if want := int64(blockSize / 2); n != want {
+		t.Fatalf("read #0: got %d bytes want %d", n, want)
+	}
+
+	// Second read (second record).
+	rr, err = r.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	n, err = io.Copy(ioutil.Discard, rr)
+	if err != io.ErrUnexpectedEOF {
+		t.Fatalf("read #1: unexpected error: %v", err)
+	}
+
+	// Third read (fourth record).
+	rr, err = r.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	n, err = io.Copy(ioutil.Discard, rr)
+	if err != nil {
+		t.Fatalf("read #2: %v", err)
+	}
+	if want := int64(blockSize-headerSize) + 2; n != want {
+		t.Fatalf("read #2: got %d bytes want %d", n, want)
+	}
+
+	if _, err := r.Next(); err != io.EOF {
+		t.Fatalf("last next: unexpected error: %v", err)
+	}
+}
+
+func TestCorrupt_CorruptedLastBlock(t *testing.T) {
+	buf := new(bytes.Buffer)
+
+	w := NewWriter(buf)
+
+	// First record.
+	ww, err := w.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := ww.Write(bytes.Repeat([]byte("0"), blockSize/2)); err != nil {
+		t.Fatalf("write #0: unexpected error: %v", err)
+	}
+
+	// Second record.
+	ww, err = w.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := ww.Write(bytes.Repeat([]byte("0"), blockSize-headerSize)); err != nil {
+		t.Fatalf("write #1: unexpected error: %v", err)
+	}
+
+	// Third record.
+	ww, err = w.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := ww.Write(bytes.Repeat([]byte("0"), (blockSize-headerSize)+1)); err != nil {
+		t.Fatalf("write #2: unexpected error: %v", err)
+	}
+
+	// Fourth record.
+	ww, err = w.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := ww.Write(bytes.Repeat([]byte("0"), (blockSize-headerSize)+2)); err != nil {
+		t.Fatalf("write #3: unexpected error: %v", err)
+	}
+
+	if err := w.Close(); err != nil {
+		t.Fatal(err)
+	}
+
+	b := buf.Bytes()
+	// Corrupting block #3.
+	for i := len(b) - 1; i > len(b)-1024; i-- {
+		b[i] = '1'
+	}
+
+	r := NewReader(bytes.NewReader(b), dropper{t}, false, true)
+
+	// First read (first record).
+	rr, err := r.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	n, err := io.Copy(ioutil.Discard, rr)
+	if err != nil {
+		t.Fatalf("read #0: %v", err)
+	}
+	if want := int64(blockSize / 2); n != want {
+		t.Fatalf("read #0: got %d bytes want %d", n, want)
+	}
+
+	// Second read (second record).
+	rr, err = r.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	n, err = io.Copy(ioutil.Discard, rr)
+	if err != nil {
+		t.Fatalf("read #1: %v", err)
+	}
+	if want := int64(blockSize - headerSize); n != want {
+		t.Fatalf("read #1: got %d bytes want %d", n, want)
+	}
+
+	// Third read (third record).
+	rr, err = r.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	n, err = io.Copy(ioutil.Discard, rr)
+	if err != nil {
+		t.Fatalf("read #2: %v", err)
+	}
+	if want := int64(blockSize-headerSize) + 1; n != want {
+		t.Fatalf("read #2: got %d bytes want %d", n, want)
+	}
+
+	// Fourth read (fourth record).
+	rr, err = r.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	n, err = io.Copy(ioutil.Discard, rr)
+	if err != io.ErrUnexpectedEOF {
+		t.Fatalf("read #3: unexpected error: %v", err)
+	}
+
+	if _, err := r.Next(); err != io.EOF {
+		t.Fatalf("last next: unexpected error: %v", err)
+	}
+}
+
+func TestCorrupt_FirstChuckLengthOverflow(t *testing.T) {
+	buf := new(bytes.Buffer)
+
+	w := NewWriter(buf)
+
+	// First record.
+	ww, err := w.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := ww.Write(bytes.Repeat([]byte("0"), blockSize/2)); err != nil {
+		t.Fatalf("write #0: unexpected error: %v", err)
+	}
+
+	// Second record.
+	ww, err = w.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := ww.Write(bytes.Repeat([]byte("0"), blockSize-headerSize)); err != nil {
+		t.Fatalf("write #1: unexpected error: %v", err)
+	}
+
+	// Third record.
+	ww, err = w.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := ww.Write(bytes.Repeat([]byte("0"), (blockSize-headerSize)+1)); err != nil {
+		t.Fatalf("write #2: unexpected error: %v", err)
+	}
+
+	if err := w.Close(); err != nil {
+		t.Fatal(err)
+	}
+
+	b := buf.Bytes()
+	// Corrupting record #1.
+	x := blockSize
+	binary.LittleEndian.PutUint16(b[x+4:], 0xffff)
+
+	r := NewReader(bytes.NewReader(b), dropper{t}, false, true)
+
+	// First read (first record).
+	rr, err := r.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	n, err := io.Copy(ioutil.Discard, rr)
+	if err != nil {
+		t.Fatalf("read #0: %v", err)
+	}
+	if want := int64(blockSize / 2); n != want {
+		t.Fatalf("read #0: got %d bytes want %d", n, want)
+	}
+
+	// Second read (second record).
+	rr, err = r.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	n, err = io.Copy(ioutil.Discard, rr)
+	if err != io.ErrUnexpectedEOF {
+		t.Fatalf("read #1: unexpected error: %v", err)
+	}
+
+	if _, err := r.Next(); err != io.EOF {
+		t.Fatalf("last next: unexpected error: %v", err)
+	}
+}
+
+func TestCorrupt_MiddleChuckLengthOverflow(t *testing.T) {
+	buf := new(bytes.Buffer)
+
+	w := NewWriter(buf)
+
+	// First record.
+	ww, err := w.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := ww.Write(bytes.Repeat([]byte("0"), blockSize/2)); err != nil {
+		t.Fatalf("write #0: unexpected error: %v", err)
+	}
+
+	// Second record.
+	ww, err = w.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := ww.Write(bytes.Repeat([]byte("0"), blockSize-headerSize)); err != nil {
+		t.Fatalf("write #1: unexpected error: %v", err)
+	}
+
+	// Third record.
+	ww, err = w.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := ww.Write(bytes.Repeat([]byte("0"), (blockSize-headerSize)+1)); err != nil {
+		t.Fatalf("write #2: unexpected error: %v", err)
+	}
+
+	if err := w.Close(); err != nil {
+		t.Fatal(err)
+	}
+
+	b := buf.Bytes()
+	// Corrupting record #1.
+	x := blockSize/2 + headerSize
+	binary.LittleEndian.PutUint16(b[x+4:], 0xffff)
+
+	r := NewReader(bytes.NewReader(b), dropper{t}, false, true)
+
+	// First read (first record).
+	rr, err := r.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	n, err := io.Copy(ioutil.Discard, rr)
+	if err != nil {
+		t.Fatalf("read #0: %v", err)
+	}
+	if want := int64(blockSize / 2); n != want {
+		t.Fatalf("read #0: got %d bytes want %d", n, want)
+	}
+
+	// Second read (third record).
+	rr, err = r.Next()
+	if err != nil {
+		t.Fatal(err)
+	}
+	n, err = io.Copy(ioutil.Discard, rr)
+	if err != nil {
+		t.Fatalf("read #1: %v", err)
+	}
+	if want := int64(blockSize-headerSize) + 1; n != want {
+		t.Fatalf("read #1: got %d bytes want %d", n, want)
+	}
+
+	if _, err := r.Next(); err != io.EOF {
+		t.Fatalf("last next: unexpected error: %v", err)
+	}
+}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/key.go

@@ -9,15 +9,30 @@ package leveldb
 import (
 	"encoding/binary"
 	"fmt"
+
+	"github.com/syndtr/goleveldb/leveldb/errors"
 )
 
-type vType int
+type ErrIkeyCorrupted struct {
+	Ikey   []byte
+	Reason string
+}
 
-func (t vType) String() string {
-	switch t {
-	case tDel:
+func (e *ErrIkeyCorrupted) Error() string {
+	return fmt.Sprintf("leveldb: iKey %q corrupted: %s", e.Ikey, e.Reason)
+}
+
+func newErrIkeyCorrupted(ikey []byte, reason string) error {
+	return errors.NewErrCorrupted(nil, &ErrIkeyCorrupted{append([]byte{}, ikey...), reason})
+}
+
+type kType int
+
+func (kt kType) String() string {
+	switch kt {
+	case ktDel:
 		return "d"
-	case tVal:
+	case ktVal:
 		return "v"
 	}
 	return "x"
@@ -26,16 +41,16 @@ func (t vType) String() string {
 // Value types encoded as the last component of internal keys.
 // Don't modify; this value are saved to disk.
 const (
-	tDel vType = iota
-	tVal
+	ktDel kType = iota
+	ktVal
 )
 
-// tSeek defines the vType that should be passed when constructing an
+// ktSeek defines the kType that should be passed when constructing an
 // internal key for seeking to a particular sequence number (since we
 // sort sequence numbers in decreasing order and the value type is
 // embedded as the low 8 bits in the sequence number in internal keys,
 // we need to use the highest-numbered ValueType, not the lowest).
-const tSeek = tVal
+const ktSeek = ktVal
 
 const (
 	// Maximum value possible for sequence number; the 8-bits are
@@ -43,7 +58,7 @@ const (
 	// 64-bit integer.
 	kMaxSeq uint64 = (uint64(1) << 56) - 1
 	// Maximum value possible for packed sequence number and type.
-	kMaxNum uint64 = (kMaxSeq << 8) | uint64(tSeek)
+	kMaxNum uint64 = (kMaxSeq << 8) | uint64(ktSeek)
 )
 
 // Maximum number encoded in bytes.
@@ -55,85 +70,73 @@ func init() {
 
 type iKey []byte
 
-func newIKey(ukey []byte, seq uint64, t vType) iKey {
-	if seq > kMaxSeq || t > tVal {
-		panic("invalid seq number or value type")
+func newIkey(ukey []byte, seq uint64, kt kType) iKey {
+	if seq > kMaxSeq {
+		panic("leveldb: invalid sequence number")
+	} else if kt > ktVal {
+		panic("leveldb: invalid type")
 	}
 
-	b := make(iKey, len(ukey)+8)
-	copy(b, ukey)
-	binary.LittleEndian.PutUint64(b[len(ukey):], (seq<<8)|uint64(t))
-	return b
+	ik := make(iKey, len(ukey)+8)
+	copy(ik, ukey)
+	binary.LittleEndian.PutUint64(ik[len(ukey):], (seq<<8)|uint64(kt))
+	return ik
 }
 
-func parseIkey(p []byte) (ukey []byte, seq uint64, t vType, ok bool) {
-	if len(p) < 8 {
-		return
+func parseIkey(ik []byte) (ukey []byte, seq uint64, kt kType, err error) {
+	if len(ik) < 8 {
+		return nil, 0, 0, newErrIkeyCorrupted(ik, "invalid length")
 	}
-	num := binary.LittleEndian.Uint64(p[len(p)-8:])
-	seq, t = uint64(num>>8), vType(num&0xff)
-	if t > tVal {
-		return
+	num := binary.LittleEndian.Uint64(ik[len(ik)-8:])
+	seq, kt = uint64(num>>8), kType(num&0xff)
+	if kt > ktVal {
+		return nil, 0, 0, newErrIkeyCorrupted(ik, "invalid type")
 	}
-	ukey = p[:len(p)-8]
-	ok = true
+	ukey = ik[:len(ik)-8]
 	return
 }
 
-func validIkey(p []byte) bool {
-	_, _, _, ok := parseIkey(p)
-	return ok
+func validIkey(ik []byte) bool {
+	_, _, _, err := parseIkey(ik)
+	return err == nil
 }
 
-func (p iKey) assert() {
-	if p == nil {
-		panic("nil iKey")
+func (ik iKey) assert() {
+	if ik == nil {
+		panic("leveldb: nil iKey")
 	}
-	if len(p) < 8 {
-		panic(fmt.Sprintf("invalid iKey %q, len=%d", []byte(p), len(p)))
+	if len(ik) < 8 {
+		panic(fmt.Sprintf("leveldb: iKey %q, len=%d: invalid length", ik, len(ik)))
 	}
 }
 
-func (p iKey) ok() bool {
-	if len(p) < 8 {
-		return false
-	}
-	_, _, ok := p.parseNum()
-	return ok
+func (ik iKey) ukey() []byte {
+	ik.assert()
+	return ik[:len(ik)-8]
 }
 
-func (p iKey) ukey() []byte {
-	p.assert()
-	return p[:len(p)-8]
+func (ik iKey) num() uint64 {
+	ik.assert()
+	return binary.LittleEndian.Uint64(ik[len(ik)-8:])
 }
 
-func (p iKey) num() uint64 {
-	p.assert()
-	return binary.LittleEndian.Uint64(p[len(p)-8:])
-}
-
-func (p iKey) parseNum() (seq uint64, t vType, ok bool) {
-	if p == nil {
-		panic("nil iKey")
+func (ik iKey) parseNum() (seq uint64, kt kType) {
+	num := ik.num()
+	seq, kt = uint64(num>>8), kType(num&0xff)
+	if kt > ktVal {
+		panic(fmt.Sprintf("leveldb: iKey %q, len=%d: invalid type %#x", ik, len(ik), kt))
 	}
-	if len(p) < 8 {
-		return
-	}
-	num := p.num()
-	seq, t = uint64(num>>8), vType(num&0xff)
-	if t > tVal {
-		return 0, 0, false
-	}
-	ok = true
 	return
 }
 
-func (p iKey) String() string {
-	if len(p) == 0 {
+func (ik iKey) String() string {
+	if ik == nil {
 		return "<nil>"
 	}
-	if seq, t, ok := p.parseNum(); ok {
-		return fmt.Sprintf("%s,%s%d", shorten(string(p.ukey())), t, seq)
+
+	if ukey, seq, kt, err := parseIkey(ik); err == nil {
+		return fmt.Sprintf("%s,%s%d", shorten(string(ukey)), kt, seq)
+	} else {
+		return "<invalid>"
 	}
-	return "<invalid>"
 }

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/key_test.go

@@ -15,8 +15,8 @@ import (
 
 var defaultIComparer = &iComparer{comparer.DefaultComparer}
 
-func ikey(key string, seq uint64, t vType) iKey {
-	return newIKey([]byte(key), uint64(seq), t)
+func ikey(key string, seq uint64, kt kType) iKey {
+	return newIkey([]byte(key), uint64(seq), kt)
 }
 
 func shortSep(a, b []byte) []byte {
@@ -37,27 +37,37 @@ func shortSuccessor(b []byte) []byte {
 	return dst
 }
 
-func testSingleKey(t *testing.T, key string, seq uint64, vt vType) {
-	ik := ikey(key, seq, vt)
+func testSingleKey(t *testing.T, key string, seq uint64, kt kType) {
+	ik := ikey(key, seq, kt)
 
 	if !bytes.Equal(ik.ukey(), []byte(key)) {
 		t.Errorf("user key does not equal, got %v, want %v", string(ik.ukey()), key)
 	}
 
-	if rseq, rt, ok := ik.parseNum(); ok {
+	rseq, rt := ik.parseNum()
+	if rseq != seq {
+		t.Errorf("seq number does not equal, got %v, want %v", rseq, seq)
+	}
+	if rt != kt {
+		t.Errorf("type does not equal, got %v, want %v", rt, kt)
+	}
+
+	if rukey, rseq, rt, kerr := parseIkey(ik); kerr == nil {
+		if !bytes.Equal(rukey, []byte(key)) {
+			t.Errorf("user key does not equal, got %v, want %v", string(ik.ukey()), key)
+		}
 		if rseq != seq {
 			t.Errorf("seq number does not equal, got %v, want %v", rseq, seq)
 		}
-
-		if rt != vt {
-			t.Errorf("type does not equal, got %v, want %v", rt, vt)
+		if rt != kt {
+			t.Errorf("type does not equal, got %v, want %v", rt, kt)
 		}
 	} else {
-		t.Error("cannot parse seq and type")
+		t.Errorf("key error: %v", kerr)
 	}
 }
 
-func TestIKey_EncodeDecode(t *testing.T) {
+func TestIkey_EncodeDecode(t *testing.T) {
 	keys := []string{"", "k", "hello", "longggggggggggggggggggggg"}
 	seqs := []uint64{
 		1, 2, 3,
@@ -67,8 +77,8 @@ func TestIKey_EncodeDecode(t *testing.T) {
 	}
 	for _, key := range keys {
 		for _, seq := range seqs {
-			testSingleKey(t, key, seq, tVal)
-			testSingleKey(t, "hello", 1, tDel)
+			testSingleKey(t, key, seq, ktVal)
+			testSingleKey(t, "hello", 1, ktDel)
 		}
 	}
 }
@@ -79,45 +89,45 @@ func assertBytes(t *testing.T, want, got []byte) {
 	}
 }
 
-func TestIKeyShortSeparator(t *testing.T) {
+func TestIkeyShortSeparator(t *testing.T) {
 	// When user keys are same
-	assertBytes(t, ikey("foo", 100, tVal),
-		shortSep(ikey("foo", 100, tVal),
-			ikey("foo", 99, tVal)))
-	assertBytes(t, ikey("foo", 100, tVal),
-		shortSep(ikey("foo", 100, tVal),
-			ikey("foo", 101, tVal)))
-	assertBytes(t, ikey("foo", 100, tVal),
-		shortSep(ikey("foo", 100, tVal),
-			ikey("foo", 100, tVal)))
-	assertBytes(t, ikey("foo", 100, tVal),
-		shortSep(ikey("foo", 100, tVal),
-			ikey("foo", 100, tDel)))
+	assertBytes(t, ikey("foo", 100, ktVal),
+		shortSep(ikey("foo", 100, ktVal),
+			ikey("foo", 99, ktVal)))
+	assertBytes(t, ikey("foo", 100, ktVal),
+		shortSep(ikey("foo", 100, ktVal),
+			ikey("foo", 101, ktVal)))
+	assertBytes(t, ikey("foo", 100, ktVal),
+		shortSep(ikey("foo", 100, ktVal),
+			ikey("foo", 100, ktVal)))
+	assertBytes(t, ikey("foo", 100, ktVal),
+		shortSep(ikey("foo", 100, ktVal),
+			ikey("foo", 100, ktDel)))
 
 	// When user keys are misordered
-	assertBytes(t, ikey("foo", 100, tVal),
-		shortSep(ikey("foo", 100, tVal),
-			ikey("bar", 99, tVal)))
+	assertBytes(t, ikey("foo", 100, ktVal),
+		shortSep(ikey("foo", 100, ktVal),
+			ikey("bar", 99, ktVal)))
 
 	// When user keys are different, but correctly ordered
-	assertBytes(t, ikey("g", uint64(kMaxSeq), tSeek),
-		shortSep(ikey("foo", 100, tVal),
-			ikey("hello", 200, tVal)))
+	assertBytes(t, ikey("g", uint64(kMaxSeq), ktSeek),
+		shortSep(ikey("foo", 100, ktVal),
+			ikey("hello", 200, ktVal)))
 
 	// When start user key is prefix of limit user key
-	assertBytes(t, ikey("foo", 100, tVal),
-		shortSep(ikey("foo", 100, tVal),
-			ikey("foobar", 200, tVal)))
+	assertBytes(t, ikey("foo", 100, ktVal),
+		shortSep(ikey("foo", 100, ktVal),
+			ikey("foobar", 200, ktVal)))
 
 	// When limit user key is prefix of start user key
-	assertBytes(t, ikey("foobar", 100, tVal),
-		shortSep(ikey("foobar", 100, tVal),
-			ikey("foo", 200, tVal)))
+	assertBytes(t, ikey("foobar", 100, ktVal),
+		shortSep(ikey("foobar", 100, ktVal),
+			ikey("foo", 200, ktVal)))
 }
 
-func TestIKeyShortestSuccessor(t *testing.T) {
-	assertBytes(t, ikey("g", uint64(kMaxSeq), tSeek),
-		shortSuccessor(ikey("foo", 100, tVal)))
-	assertBytes(t, ikey("\xff\xff", 100, tVal),
-		shortSuccessor(ikey("\xff\xff", 100, tVal)))
+func TestIkeyShortestSuccessor(t *testing.T) {
+	assertBytes(t, ikey("g", uint64(kMaxSeq), ktSeek),
+		shortSuccessor(ikey("foo", 100, ktVal)))
+	assertBytes(t, ikey("\xff\xff", 100, ktVal),
+		shortSuccessor(ikey("\xff\xff", 100, ktVal)))
 }

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/memdb/memdb.go

@@ -12,12 +12,14 @@ import (
 	"sync"
 
 	"github.com/syndtr/goleveldb/leveldb/comparer"
+	"github.com/syndtr/goleveldb/leveldb/errors"
 	"github.com/syndtr/goleveldb/leveldb/iterator"
 	"github.com/syndtr/goleveldb/leveldb/util"
 )
 
 var (
-	ErrNotFound = util.ErrNotFound
+	ErrNotFound     = errors.ErrNotFound
+	ErrIterReleased = errors.New("leveldb/memdb: iterator released")
 )
 
 const tMaxHeight = 12
@@ -29,6 +31,7 @@ type dbIter struct {
 	node       int
 	forward    bool
 	key, value []byte
+	err        error
 }
 
 func (i *dbIter) fill(checkStart, checkLimit bool) bool {
@@ -59,6 +62,11 @@ func (i *dbIter) Valid() bool {
 }
 
 func (i *dbIter) First() bool {
+	if i.Released() {
+		i.err = ErrIterReleased
+		return false
+	}
+
 	i.forward = true
 	i.p.mu.RLock()
 	defer i.p.mu.RUnlock()
@@ -71,9 +79,11 @@ func (i *dbIter) First() bool {
 }
 
 func (i *dbIter) Last() bool {
-	if i.p == nil {
+	if i.Released() {
+		i.err = ErrIterReleased
 		return false
 	}
+
 	i.forward = false
 	i.p.mu.RLock()
 	defer i.p.mu.RUnlock()
@@ -86,9 +96,11 @@ func (i *dbIter) Last() bool {
 }
 
 func (i *dbIter) Seek(key []byte) bool {
-	if i.p == nil {
+	if i.Released() {
+		i.err = ErrIterReleased
 		return false
 	}
+
 	i.forward = true
 	i.p.mu.RLock()
 	defer i.p.mu.RUnlock()
@@ -100,9 +112,11 @@ func (i *dbIter) Seek(key []byte) bool {
 }
 
 func (i *dbIter) Next() bool {
-	if i.p == nil {
+	if i.Released() {
+		i.err = ErrIterReleased
 		return false
 	}
+
 	if i.node == 0 {
 		if !i.forward {
 			return i.First()
@@ -117,9 +131,11 @@ func (i *dbIter) Next() bool {
 }
 
 func (i *dbIter) Prev() bool {
-	if i.p == nil {
+	if i.Released() {
+		i.err = ErrIterReleased
 		return false
 	}
+
 	if i.node == 0 {
 		if i.forward {
 			return i.Last()
@@ -141,10 +157,10 @@ func (i *dbIter) Value() []byte {
 	return i.value
 }
 
-func (i *dbIter) Error() error { return nil }
+func (i *dbIter) Error() error { return i.err }
 
 func (i *dbIter) Release() {
-	if i.p != nil {
+	if !i.Released() {
 		i.p = nil
 		i.node = 0
 		i.key = nil
@@ -437,6 +453,8 @@ func (p *DB) Reset() {
 // New creates a new initalized in-memory key/value DB. The capacity
 // is the initial key/value buffer capacity. The capacity is advisory,
 // not enforced.
+//
+// The returned DB instance is goroutine-safe.
 func New(cmp comparer.BasicComparer, capacity int) *DB {
 	p := &DB{
 		cmp:       cmp,

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/memdb/memdb_test.go

@@ -129,7 +129,7 @@ var _ = testutil.Defer(func() {
 				}
 
 				return db
-			})
+			}, nil, nil)
 		})
 	})
 })

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/opt/options.go

@@ -11,6 +11,7 @@ import (
 	"github.com/syndtr/goleveldb/leveldb/cache"
 	"github.com/syndtr/goleveldb/leveldb/comparer"
 	"github.com/syndtr/goleveldb/leveldb/filter"
+	"math"
 )
 
 const (
@@ -20,20 +21,38 @@ const (
 )
 
 const (
-	DefaultBlockCacheSize       = 8 * MiB
-	DefaultBlockRestartInterval = 16
-	DefaultBlockSize            = 4 * KiB
-	DefaultCompressionType      = SnappyCompression
-	DefaultMaxOpenFiles         = 1000
-	DefaultWriteBuffer          = 4 * MiB
+	DefaultBlockCacheSize                = 8 * MiB
+	DefaultBlockRestartInterval          = 16
+	DefaultBlockSize                     = 4 * KiB
+	DefaultCompactionExpandLimitFactor   = 25
+	DefaultCompactionGPOverlapsFactor    = 10
+	DefaultCompactionL0Trigger           = 4
+	DefaultCompactionSourceLimitFactor   = 1
+	DefaultCompactionTableSize           = 2 * MiB
+	DefaultCompactionTableSizeMultiplier = 1.0
+	DefaultCompactionTotalSize           = 10 * MiB
+	DefaultCompactionTotalSizeMultiplier = 10.0
+	DefaultCompressionType               = SnappyCompression
+	DefaultCachedOpenFiles               = 500
+	DefaultMaxMemCompationLevel          = 2
+	DefaultNumLevel                      = 7
+	DefaultWriteBuffer                   = 4 * MiB
+	DefaultWriteL0PauseTrigger           = 12
+	DefaultWriteL0SlowdownTrigger        = 8
 )
 
 type noCache struct{}
 
-func (noCache) SetCapacity(capacity int)               {}
-func (noCache) GetNamespace(id uint64) cache.Namespace { return nil }
-func (noCache) Purge(fin cache.PurgeFin)               {}
-func (noCache) Zap(closed bool)                        {}
+func (noCache) SetCapacity(capacity int)                     {}
+func (noCache) Capacity() int                                { return 0 }
+func (noCache) Used() int                                    { return 0 }
+func (noCache) Size() int                                    { return 0 }
+func (noCache) NumObjects() int                              { return 0 }
+func (noCache) GetNamespace(id uint64) cache.Namespace       { return nil }
+func (noCache) PurgeNamespace(id uint64, fin cache.PurgeFin) {}
+func (noCache) ZapNamespace(id uint64)                       {}
+func (noCache) Purge(fin cache.PurgeFin)                     {}
+func (noCache) Zap()                                         {}
 
 var NoCache cache.Cache = noCache{}
 
@@ -59,34 +78,47 @@ const (
 	nCompression
 )
 
-// Strict is the DB strict level.
+// Strict is the DB 'strict level'.
 type Strict uint
 
 const (
 	// If present then a corrupted or invalid chunk or block in manifest
-	// journal will cause an error istead of being dropped.
+	// journal will cause an error instead of being dropped.
+	// This will prevent database with corrupted manifest to be opened.
 	StrictManifest Strict = 1 << iota
 
-	// If present then a corrupted or invalid chunk or block in journal
-	// will cause an error istead of being dropped.
-	StrictJournal
-
 	// If present then journal chunk checksum will be verified.
 	StrictJournalChecksum
 
-	// If present then an invalid key/value pair will cause an error
-	// instead of being skipped.
-	StrictIterator
+	// If present then a corrupted or invalid chunk or block in journal
+	// will cause an error instead of being dropped.
+	// This will prevent database with corrupted journal to be opened.
+	StrictJournal
 
 	// If present then 'sorted table' block checksum will be verified.
+	// This has effect on both 'read operation' and compaction.
 	StrictBlockChecksum
 
+	// If present then a corrupted 'sorted table' will fails compaction.
+	// The database will enter read-only mode.
+	StrictCompaction
+
+	// If present then a corrupted 'sorted table' will halts 'read operation'.
+	StrictReader
+
+	// If present then leveldb.Recover will drop corrupted 'sorted table'.
+	StrictRecovery
+
+	// This only applicable for ReadOptions, if present then this ReadOptions
+	// 'strict level' will override global ones.
+	StrictOverride
+
 	// StrictAll enables all strict flags.
-	StrictAll = StrictManifest | StrictJournal | StrictJournalChecksum | StrictIterator | StrictBlockChecksum
+	StrictAll = StrictManifest | StrictJournalChecksum | StrictJournal | StrictBlockChecksum | StrictCompaction | StrictReader
 
 	// DefaultStrict is the default strict flags. Specify any strict flags
 	// will override default strict flags as whole (i.e. not OR'ed).
-	DefaultStrict = StrictJournalChecksum | StrictBlockChecksum
+	DefaultStrict = StrictJournalChecksum | StrictBlockChecksum | StrictCompaction | StrictReader
 
 	// NoStrict disables all strict flags. Override default strict flags.
 	NoStrict = ^StrictAll
@@ -119,6 +151,80 @@ type Options struct {
 	// The default value is 4KiB.
 	BlockSize int
 
+	// CachedOpenFiles defines number of open files to kept around when not
+	// in-use, the counting includes still in-use files.
+	// Set this to negative value to disable caching.
+	//
+	// The default value is 500.
+	CachedOpenFiles int
+
+	// CompactionExpandLimitFactor limits compaction size after expanded.
+	// This will be multiplied by table size limit at compaction target level.
+	//
+	// The default value is 25.
+	CompactionExpandLimitFactor int
+
+	// CompactionGPOverlapsFactor limits overlaps in grandparent (Level + 2) that a
+	// single 'sorted table' generates.
+	// This will be multiplied by table size limit at grandparent level.
+	//
+	// The default value is 10.
+	CompactionGPOverlapsFactor int
+
+	// CompactionL0Trigger defines number of 'sorted table' at level-0 that will
+	// trigger compaction.
+	//
+	// The default value is 4.
+	CompactionL0Trigger int
+
+	// CompactionSourceLimitFactor limits compaction source size. This doesn't apply to
+	// level-0.
+	// This will be multiplied by table size limit at compaction target level.
+	//
+	// The default value is 1.
+	CompactionSourceLimitFactor int
+
+	// CompactionTableSize limits size of 'sorted table' that compaction generates.
+	// The limits for each level will be calculated as:
+	//   CompactionTableSize * (CompactionTableSizeMultiplier ^ Level)
+	// The multiplier for each level can also fine-tuned using CompactionTableSizeMultiplierPerLevel.
+	//
+	// The default value is 2MiB.
+	CompactionTableSize int
+
+	// CompactionTableSizeMultiplier defines multiplier for CompactionTableSize.
+	//
+	// The default value is 1.
+	CompactionTableSizeMultiplier float64
+
+	// CompactionTableSizeMultiplierPerLevel defines per-level multiplier for
+	// CompactionTableSize.
+	// Use zero to skip a level.
+	//
+	// The default value is nil.
+	CompactionTableSizeMultiplierPerLevel []float64
+
+	// CompactionTotalSize limits total size of 'sorted table' for each level.
+	// The limits for each level will be calculated as:
+	//   CompactionTotalSize * (CompactionTotalSizeMultiplier ^ Level)
+	// The multiplier for each level can also fine-tuned using
+	// CompactionTotalSizeMultiplierPerLevel.
+	//
+	// The default value is 10MiB.
+	CompactionTotalSize int
+
+	// CompactionTotalSizeMultiplier defines multiplier for CompactionTotalSize.
+	//
+	// The default value is 10.
+	CompactionTotalSizeMultiplier float64
+
+	// CompactionTotalSizeMultiplierPerLevel defines per-level multiplier for
+	// CompactionTotalSize.
+	// Use zero to skip a level.
+	//
+	// The default value is nil.
+	CompactionTotalSizeMultiplierPerLevel []float64
+
 	// Comparer defines a total ordering over the space of []byte keys: a 'less
 	// than' relationship. The same comparison algorithm must be used for reads
 	// and writes over the lifetime of the DB.
@@ -131,6 +237,11 @@ type Options struct {
 	// The default value (DefaultCompression) uses snappy compression.
 	Compression Compression
 
+	// DisableCompactionBackoff allows disable compaction retry backoff.
+	//
+	// The default value is false.
+	DisableCompactionBackoff bool
+
 	// ErrorIfExist defines whether an error should returned if the DB already
 	// exist.
 	//
@@ -159,12 +270,18 @@ type Options struct {
 	// The default value is nil.
 	Filter filter.Filter
 
-	// MaxOpenFiles defines maximum number of open files to kept around
-	// (cached). This is not an hard limit, actual open files may exceed
-	// the defined value.
+	// MaxMemCompationLevel defines maximum level a newly compacted 'memdb'
+	// will be pushed into if doesn't creates overlap. This should less than
+	// NumLevel. Use -1 for level-0.
 	//
-	// The default value is 1000.
-	MaxOpenFiles int
+	// The default is 2.
+	MaxMemCompationLevel int
+
+	// NumLevel defines number of database level. The level shouldn't changed
+	// between opens, or the database will panic.
+	//
+	// The default is 7.
+	NumLevel int
 
 	// Strict defines the DB strict level.
 	Strict Strict
@@ -177,6 +294,18 @@ type Options struct {
 	//
 	// The default value is 4MiB.
 	WriteBuffer int
+
+	// WriteL0StopTrigger defines number of 'sorted table' at level-0 that will
+	// pause write.
+	//
+	// The default value is 12.
+	WriteL0PauseTrigger int
+
+	// WriteL0SlowdownTrigger defines number of 'sorted table' at level-0 that
+	// will trigger write slowdown.
+	//
+	// The default value is 8.
+	WriteL0SlowdownTrigger int
 }
 
 func (o *Options) GetAltFilters() []filter.Filter {
@@ -207,6 +336,88 @@ func (o *Options) GetBlockSize() int {
 	return o.BlockSize
 }
 
+func (o *Options) GetCachedOpenFiles() int {
+	if o == nil || o.CachedOpenFiles == 0 {
+		return DefaultCachedOpenFiles
+	} else if o.CachedOpenFiles < 0 {
+		return 0
+	}
+	return o.CachedOpenFiles
+}
+
+func (o *Options) GetCompactionExpandLimit(level int) int {
+	factor := DefaultCompactionExpandLimitFactor
+	if o != nil && o.CompactionExpandLimitFactor > 0 {
+		factor = o.CompactionExpandLimitFactor
+	}
+	return o.GetCompactionTableSize(level+1) * factor
+}
+
+func (o *Options) GetCompactionGPOverlaps(level int) int {
+	factor := DefaultCompactionGPOverlapsFactor
+	if o != nil && o.CompactionGPOverlapsFactor > 0 {
+		factor = o.CompactionGPOverlapsFactor
+	}
+	return o.GetCompactionTableSize(level+2) * factor
+}
+
+func (o *Options) GetCompactionL0Trigger() int {
+	if o == nil || o.CompactionL0Trigger == 0 {
+		return DefaultCompactionL0Trigger
+	}
+	return o.CompactionL0Trigger
+}
+
+func (o *Options) GetCompactionSourceLimit(level int) int {
+	factor := DefaultCompactionSourceLimitFactor
+	if o != nil && o.CompactionSourceLimitFactor > 0 {
+		factor = o.CompactionSourceLimitFactor
+	}
+	return o.GetCompactionTableSize(level+1) * factor
+}
+
+func (o *Options) GetCompactionTableSize(level int) int {
+	var (
+		base = DefaultCompactionTableSize
+		mult float64
+	)
+	if o != nil {
+		if o.CompactionTableSize > 0 {
+			base = o.CompactionTableSize
+		}
+		if len(o.CompactionTableSizeMultiplierPerLevel) > level && o.CompactionTableSizeMultiplierPerLevel[level] > 0 {
+			mult = o.CompactionTableSizeMultiplierPerLevel[level]
+		} else if o.CompactionTableSizeMultiplier > 0 {
+			mult = math.Pow(o.CompactionTableSizeMultiplier, float64(level))
+		}
+	}
+	if mult == 0 {
+		mult = math.Pow(DefaultCompactionTableSizeMultiplier, float64(level))
+	}
+	return int(float64(base) * mult)
+}
+
+func (o *Options) GetCompactionTotalSize(level int) int64 {
+	var (
+		base = DefaultCompactionTotalSize
+		mult float64
+	)
+	if o != nil {
+		if o.CompactionTotalSize > 0 {
+			base = o.CompactionTotalSize
+		}
+		if len(o.CompactionTotalSizeMultiplierPerLevel) > level && o.CompactionTotalSizeMultiplierPerLevel[level] > 0 {
+			mult = o.CompactionTotalSizeMultiplierPerLevel[level]
+		} else if o.CompactionTotalSizeMultiplier > 0 {
+			mult = math.Pow(o.CompactionTotalSizeMultiplier, float64(level))
+		}
+	}
+	if mult == 0 {
+		mult = math.Pow(DefaultCompactionTotalSizeMultiplier, float64(level))
+	}
+	return int64(float64(base) * mult)
+}
+
 func (o *Options) GetComparer() comparer.Comparer {
 	if o == nil || o.Comparer == nil {
 		return comparer.DefaultComparer
@@ -221,6 +432,13 @@ func (o *Options) GetCompression() Compression {
 	return o.Compression
 }
 
+func (o *Options) GetDisableCompactionBackoff() bool {
+	if o == nil {
+		return false
+	}
+	return o.DisableCompactionBackoff
+}
+
 func (o *Options) GetErrorIfExist() bool {
 	if o == nil {
 		return false
@@ -242,11 +460,26 @@ func (o *Options) GetFilter() filter.Filter {
 	return o.Filter
 }
 
-func (o *Options) GetMaxOpenFiles() int {
-	if o == nil || o.MaxOpenFiles <= 0 {
-		return DefaultMaxOpenFiles
+func (o *Options) GetMaxMemCompationLevel() int {
+	level := DefaultMaxMemCompationLevel
+	if o != nil {
+		if o.MaxMemCompationLevel > 0 {
+			level = o.MaxMemCompationLevel
+		} else if o.MaxMemCompationLevel == -1 {
+			level = 0
+		}
 	}
-	return o.MaxOpenFiles
+	if level >= o.GetNumLevel() {
+		return o.GetNumLevel() - 1
+	}
+	return level
+}
+
+func (o *Options) GetNumLevel() int {
+	if o == nil || o.NumLevel <= 0 {
+		return DefaultNumLevel
+	}
+	return o.NumLevel
 }
 
 func (o *Options) GetStrict(strict Strict) bool {
@@ -263,6 +496,20 @@ func (o *Options) GetWriteBuffer() int {
 	return o.WriteBuffer
 }
 
+func (o *Options) GetWriteL0PauseTrigger() int {
+	if o == nil || o.WriteL0PauseTrigger == 0 {
+		return DefaultWriteL0PauseTrigger
+	}
+	return o.WriteL0PauseTrigger
+}
+
+func (o *Options) GetWriteL0SlowdownTrigger() int {
+	if o == nil || o.WriteL0SlowdownTrigger == 0 {
+		return DefaultWriteL0SlowdownTrigger
+	}
+	return o.WriteL0SlowdownTrigger
+}
+
 // ReadOptions holds the optional parameters for 'read operation'. The
 // 'read operation' includes Get, Find and NewIterator.
 type ReadOptions struct {
@@ -273,8 +520,8 @@ type ReadOptions struct {
 	// The default value is false.
 	DontFillCache bool
 
-	// Strict overrides global DB strict level. Only StrictIterator and
-	// StrictBlockChecksum that does have effects here.
+	// Strict will be OR'ed with global DB 'strict level' unless StrictOverride
+	// is present. Currently only StrictReader that has effect here.
 	Strict Strict
 }
 
@@ -316,3 +563,11 @@ func (wo *WriteOptions) GetSync() bool {
 	}
 	return wo.Sync
 }
+
+func GetStrict(o *Options, ro *ReadOptions, strict Strict) bool {
+	if ro.GetStrict(StrictOverride) {
+		return ro.GetStrict(strict)
+	} else {
+		return o.GetStrict(strict) || ro.GetStrict(strict)
+	}
+}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/options.go

@@ -12,30 +12,86 @@ import (
 	"github.com/syndtr/goleveldb/leveldb/opt"
 )
 
-func (s *session) setOptions(o *opt.Options) {
-	s.o = &opt.Options{}
+func dupOptions(o *opt.Options) *opt.Options {
+	newo := &opt.Options{}
 	if o != nil {
-		*s.o = *o
+		*newo = *o
 	}
+	return newo
+}
+
+func (s *session) setOptions(o *opt.Options) {
+	no := dupOptions(o)
 	// Alternative filters.
 	if filters := o.GetAltFilters(); len(filters) > 0 {
-		s.o.AltFilters = make([]filter.Filter, len(filters))
+		no.AltFilters = make([]filter.Filter, len(filters))
 		for i, filter := range filters {
-			s.o.AltFilters[i] = &iFilter{filter}
+			no.AltFilters[i] = &iFilter{filter}
 		}
 	}
 	// Block cache.
 	switch o.GetBlockCache() {
 	case nil:
-		s.o.BlockCache = cache.NewLRUCache(opt.DefaultBlockCacheSize)
+		no.BlockCache = cache.NewLRUCache(opt.DefaultBlockCacheSize)
 	case opt.NoCache:
-		s.o.BlockCache = nil
+		no.BlockCache = nil
 	}
 	// Comparer.
 	s.icmp = &iComparer{o.GetComparer()}
-	s.o.Comparer = s.icmp
+	no.Comparer = s.icmp
 	// Filter.
 	if filter := o.GetFilter(); filter != nil {
-		s.o.Filter = &iFilter{filter}
+		no.Filter = &iFilter{filter}
+	}
+
+	s.o = &cachedOptions{Options: no}
+	s.o.cache()
+}
+
+type cachedOptions struct {
+	*opt.Options
+
+	compactionExpandLimit []int
+	compactionGPOverlaps  []int
+	compactionSourceLimit []int
+	compactionTableSize   []int
+	compactionTotalSize   []int64
+}
+
+func (co *cachedOptions) cache() {
+	numLevel := co.Options.GetNumLevel()
+
+	co.compactionExpandLimit = make([]int, numLevel)
+	co.compactionGPOverlaps = make([]int, numLevel)
+	co.compactionSourceLimit = make([]int, numLevel)
+	co.compactionTableSize = make([]int, numLevel)
+	co.compactionTotalSize = make([]int64, numLevel)
+
+	for level := 0; level < numLevel; level++ {
+		co.compactionExpandLimit[level] = co.Options.GetCompactionExpandLimit(level)
+		co.compactionGPOverlaps[level] = co.Options.GetCompactionGPOverlaps(level)
+		co.compactionSourceLimit[level] = co.Options.GetCompactionSourceLimit(level)
+		co.compactionTableSize[level] = co.Options.GetCompactionTableSize(level)
+		co.compactionTotalSize[level] = co.Options.GetCompactionTotalSize(level)
 	}
 }
+
+func (co *cachedOptions) GetCompactionExpandLimit(level int) int {
+	return co.compactionExpandLimit[level]
+}
+
+func (co *cachedOptions) GetCompactionGPOverlaps(level int) int {
+	return co.compactionGPOverlaps[level]
+}
+
+func (co *cachedOptions) GetCompactionSourceLimit(level int) int {
+	return co.compactionSourceLimit[level]
+}
+
+func (co *cachedOptions) GetCompactionTableSize(level int) int {
+	return co.compactionTableSize[level]
+}
+
+func (co *cachedOptions) GetCompactionTotalSize(level int) int64 {
+	return co.compactionTotalSize[level]
+}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/session.go

@@ -7,12 +7,13 @@
 package leveldb
 
 import (
-	"errors"
+	"fmt"
 	"io"
 	"os"
 	"sync"
 	"sync/atomic"
 
+	"github.com/syndtr/goleveldb/leveldb/errors"
 	"github.com/syndtr/goleveldb/leveldb/iterator"
 	"github.com/syndtr/goleveldb/leveldb/journal"
 	"github.com/syndtr/goleveldb/leveldb/opt"
@@ -20,18 +21,31 @@ import (
 	"github.com/syndtr/goleveldb/leveldb/util"
 )
 
+type ErrManifestCorrupted struct {
+	Field  string
+	Reason string
+}
+
+func (e *ErrManifestCorrupted) Error() string {
+	return fmt.Sprintf("leveldb: manifest corrupted (field '%s'): %s", e.Field, e.Reason)
+}
+
+func newErrManifestCorrupted(f storage.File, field, reason string) error {
+	return errors.NewErrCorrupted(f, &ErrManifestCorrupted{field, reason})
+}
+
 // session represent a persistent database session.
 type session struct {
 	// Need 64-bit alignment.
-	stFileNum        uint64 // current unused file number
+	stNextFileNum    uint64 // current unused file number
 	stJournalNum     uint64 // current journal file number; need external synchronization
 	stPrevJournalNum uint64 // prev journal file number; no longer used; for compatibility with older version of leveldb
-	stSeq            uint64 // last mem compacted seq; need external synchronization
+	stSeqNum         uint64 // last mem compacted seq; need external synchronization
 	stTempFileNum    uint64
 
 	stor     storage.Storage
 	storLock util.Releaser
-	o        *opt.Options
+	o        *cachedOptions
 	icmp     *iComparer
 	tops     *tOps
 
@@ -39,11 +53,12 @@ type session struct {
 	manifestWriter storage.Writer
 	manifestFile   storage.File
 
-	stCPtrs   [kNumLevels]iKey // compact pointers; need external synchronization
-	stVersion *version         // current version
-	vmu       sync.Mutex
+	stCompPtrs []iKey   // compaction pointers; need external synchronization
+	stVersion  *version // current version
+	vmu        sync.Mutex
 }
 
+// Creates new initialized session instance.
 func newSession(stor storage.Storage, o *opt.Options) (s *session, err error) {
 	if stor == nil {
 		return nil, os.ErrInvalid
@@ -53,13 +68,14 @@ func newSession(stor storage.Storage, o *opt.Options) (s *session, err error) {
 		return
 	}
 	s = &session{
-		stor:     stor,
-		storLock: storLock,
+		stor:       stor,
+		storLock:   storLock,
+		stCompPtrs: make([]iKey, o.GetNumLevel()),
 	}
 	s.setOptions(o)
-	s.tops = newTableOps(s, s.o.GetMaxOpenFiles())
-	s.setVersion(&version{s: s})
-	s.log("log@legend F·NumFile S·FileSize N·Entry C·BadEntry B·BadBlock D·DeletedEntry L·Level Q·SeqNum T·TimeElapsed")
+	s.tops = newTableOps(s, s.o.GetCachedOpenFiles())
+	s.setVersion(newVersion(s))
+	s.log("log@legend F·NumFile S·FileSize N·Entry C·BadEntry B·BadBlock Ke·KeyError D·DroppedEntry L·Level Q·SeqNum T·TimeElapsed")
 	return
 }
 
@@ -81,6 +97,7 @@ func (s *session) close() {
 	s.stVersion = nil
 }
 
+// Release session lock.
 func (s *session) release() {
 	s.storLock.Release()
 }
@@ -98,26 +115,26 @@ func (s *session) recover() (err error) {
 			// Don't return os.ErrNotExist if the underlying storage contains
 			// other files that belong to LevelDB. So the DB won't get trashed.
 			if files, _ := s.stor.GetFiles(storage.TypeAll); len(files) > 0 {
-				err = ErrCorrupted{Type: CorruptedManifest, Err: errors.New("leveldb: manifest file missing")}
+				err = &errors.ErrCorrupted{File: &storage.FileInfo{Type: storage.TypeManifest}, Err: &errors.ErrMissingFiles{}}
 			}
 		}
 	}()
 
-	file, err := s.stor.GetManifest()
+	m, err := s.stor.GetManifest()
 	if err != nil {
 		return
 	}
 
-	reader, err := file.Open()
+	reader, err := m.Open()
 	if err != nil {
 		return
 	}
 	defer reader.Close()
 	strict := s.o.GetStrict(opt.StrictManifest)
-	jr := journal.NewReader(reader, dropper{s, file}, strict, true)
+	jr := journal.NewReader(reader, dropper{s, m}, strict, true)
 
-	staging := s.version_NB().newStaging()
-	rec := &sessionRecord{}
+	staging := s.stVersion.newStaging()
+	rec := &sessionRecord{numLevel: s.o.GetNumLevel()}
 	for {
 		var r io.Reader
 		r, err = jr.Next()
@@ -126,51 +143,57 @@ func (s *session) recover() (err error) {
 				err = nil
 				break
 			}
-			return
+			return errors.SetFile(err, m)
 		}
 
 		err = rec.decode(r)
 		if err == nil {
 			// save compact pointers
-			for _, rp := range rec.compactionPointers {
-				s.stCPtrs[rp.level] = iKey(rp.key)
+			for _, r := range rec.compPtrs {
+				s.stCompPtrs[r.level] = iKey(r.ikey)
 			}
 			// commit record to version staging
 			staging.commit(rec)
-		} else if strict {
-			return ErrCorrupted{Type: CorruptedManifest, Err: err}
 		} else {
-			s.logf("manifest error: %v (skipped)", err)
+			err = errors.SetFile(err, m)
+			if strict || !errors.IsCorrupted(err) {
+				return
+			} else {
+				s.logf("manifest error: %v (skipped)", errors.SetFile(err, m))
+			}
 		}
-		rec.resetCompactionPointers()
+		rec.resetCompPtrs()
 		rec.resetAddedTables()
 		rec.resetDeletedTables()
 	}
 
 	switch {
 	case !rec.has(recComparer):
-		return ErrCorrupted{Type: CorruptedManifest, Err: errors.New("leveldb: manifest missing comparer name")}
+		return newErrManifestCorrupted(m, "comparer", "missing")
 	case rec.comparer != s.icmp.uName():
-		return ErrCorrupted{Type: CorruptedManifest, Err: errors.New("leveldb: comparer mismatch, " + "want '" + s.icmp.uName() + "', " + "got '" + rec.comparer + "'")}
-	case !rec.has(recNextNum):
-		return ErrCorrupted{Type: CorruptedManifest, Err: errors.New("leveldb: manifest missing next file number")}
+		return newErrManifestCorrupted(m, "comparer", fmt.Sprintf("mismatch: want '%s', got '%s'", s.icmp.uName(), rec.comparer))
+	case !rec.has(recNextFileNum):
+		return newErrManifestCorrupted(m, "next-file-num", "missing")
 	case !rec.has(recJournalNum):
-		return ErrCorrupted{Type: CorruptedManifest, Err: errors.New("leveldb: manifest missing journal file number")}
-	case !rec.has(recSeq):
-		return ErrCorrupted{Type: CorruptedManifest, Err: errors.New("leveldb: manifest missing seq number")}
+		return newErrManifestCorrupted(m, "journal-file-num", "missing")
+	case !rec.has(recSeqNum):
+		return newErrManifestCorrupted(m, "seq-num", "missing")
 	}
 
-	s.manifestFile = file
+	s.manifestFile = m
 	s.setVersion(staging.finish())
-	s.setFileNum(rec.nextNum)
+	s.setNextFileNum(rec.nextFileNum)
 	s.recordCommited(rec)
 	return nil
 }
 
 // Commit session; need external synchronization.
 func (s *session) commit(r *sessionRecord) (err error) {
+	v := s.version()
+	defer v.release()
+
 	// spawn new version based on current version
-	nv := s.version_NB().spawn(r)
+	nv := v.spawn(r)
 
 	if s.manifest == nil {
 		// manifest journal writer not yet created, create one
@@ -189,22 +212,22 @@ func (s *session) commit(r *sessionRecord) (err error) {
 
 // Pick a compaction based on current state; need external synchronization.
 func (s *session) pickCompaction() *compaction {
-	v := s.version_NB()
+	v := s.version()
 
 	var level int
 	var t0 tFiles
 	if v.cScore >= 1 {
 		level = v.cLevel
-		cp := s.stCPtrs[level]
-		tt := v.tables[level]
-		for _, t := range tt {
-			if cp == nil || s.icmp.Compare(t.max, cp) > 0 {
+		cptr := s.stCompPtrs[level]
+		tables := v.tables[level]
+		for _, t := range tables {
+			if cptr == nil || s.icmp.Compare(t.imax, cptr) > 0 {
 				t0 = append(t0, t)
 				break
 			}
 		}
 		if len(t0) == 0 {
-			t0 = append(t0, tt[0])
+			t0 = append(t0, tables[0])
 		}
 	} else {
 		if p := atomic.LoadPointer(&v.cSeek); p != nil {
@@ -212,29 +235,21 @@ func (s *session) pickCompaction() *compaction {
 			level = ts.level
 			t0 = append(t0, ts.table)
 		} else {
+			v.release()
 			return nil
 		}
 	}
 
-	c := &compaction{s: s, version: v, level: level}
-	if level == 0 {
-		min, max := t0.getRange(s.icmp)
-		t0 = nil
-		v.tables[0].getOverlaps(min.ukey(), max.ukey(), &t0, false, s.icmp.ucmp)
-	}
-
-	c.tables[0] = t0
-	c.expand()
-	return c
+	return newCompaction(s, v, level, t0)
 }
 
 // Create compaction from given level and range; need external synchronization.
-func (s *session) getCompactionRange(level int, min, max []byte) *compaction {
-	v := s.version_NB()
+func (s *session) getCompactionRange(level int, umin, umax []byte) *compaction {
+	v := s.version()
 
-	var t0 tFiles
-	v.tables[level].getOverlaps(min, max, &t0, level != 0, s.icmp.ucmp)
+	t0 := v.tables[level].getOverlaps(nil, s.icmp, umin, umax, level == 0)
 	if len(t0) == 0 {
+		v.release()
 		return nil
 	}
 
@@ -243,7 +258,7 @@ func (s *session) getCompactionRange(level int, min, max []byte) *compaction {
 	// and we must not pick one file and drop another older file if the
 	// two files overlap.
 	if level > 0 {
-		limit := uint64(kMaxTableSize)
+		limit := uint64(v.s.o.GetCompactionSourceLimit(level))
 		total := uint64(0)
 		for i, t := range t0 {
 			total += t.size
@@ -255,90 +270,124 @@ func (s *session) getCompactionRange(level int, min, max []byte) *compaction {
 		}
 	}
 
-	c := &compaction{s: s, version: v, level: level}
-	c.tables[0] = t0
+	return newCompaction(s, v, level, t0)
+}
+
+func newCompaction(s *session, v *version, level int, t0 tFiles) *compaction {
+	c := &compaction{
+		s:             s,
+		v:             v,
+		level:         level,
+		tables:        [2]tFiles{t0, nil},
+		maxGPOverlaps: uint64(s.o.GetCompactionGPOverlaps(level)),
+		tPtrs:         make([]int, s.o.GetNumLevel()),
+	}
 	c.expand()
+	c.save()
 	return c
 }
 
-// compaction represent a compaction state
+// compaction represent a compaction state.
 type compaction struct {
-	s       *session
-	version *version
+	s *session
+	v *version
 
-	level  int
-	tables [2]tFiles
+	level         int
+	tables        [2]tFiles
+	maxGPOverlaps uint64
 
-	gp              tFiles
-	gpidx           int
-	seenKey         bool
-	overlappedBytes uint64
-	min, max        iKey
+	gp                tFiles
+	gpi               int
+	seenKey           bool
+	gpOverlappedBytes uint64
+	imin, imax        iKey
+	tPtrs             []int
+	released          bool
 
-	tPtrs [kNumLevels]int
+	snapGPI               int
+	snapSeenKey           bool
+	snapGPOverlappedBytes uint64
+	snapTPtrs             []int
+}
+
+func (c *compaction) save() {
+	c.snapGPI = c.gpi
+	c.snapSeenKey = c.seenKey
+	c.snapGPOverlappedBytes = c.gpOverlappedBytes
+	c.snapTPtrs = append(c.snapTPtrs[:0], c.tPtrs...)
+}
+
+func (c *compaction) restore() {
+	c.gpi = c.snapGPI
+	c.seenKey = c.snapSeenKey
+	c.gpOverlappedBytes = c.snapGPOverlappedBytes
+	c.tPtrs = append(c.tPtrs[:0], c.snapTPtrs...)
+}
+
+func (c *compaction) release() {
+	if !c.released {
+		c.released = true
+		c.v.release()
+	}
 }
 
 // Expand compacted tables; need external synchronization.
 func (c *compaction) expand() {
-	s := c.s
-	v := c.version
-
-	level := c.level
-	vt0, vt1 := v.tables[level], v.tables[level+1]
+	limit := uint64(c.s.o.GetCompactionExpandLimit(c.level))
+	vt0, vt1 := c.v.tables[c.level], c.v.tables[c.level+1]
 
 	t0, t1 := c.tables[0], c.tables[1]
-	min, max := t0.getRange(s.icmp)
-	vt1.getOverlaps(min.ukey(), max.ukey(), &t1, true, s.icmp.ucmp)
-
-	// Get entire range covered by compaction
-	amin, amax := append(t0, t1...).getRange(s.icmp)
+	imin, imax := t0.getRange(c.s.icmp)
+	// We expand t0 here just incase ukey hop across tables.
+	t0 = vt0.getOverlaps(t0, c.s.icmp, imin.ukey(), imax.ukey(), c.level == 0)
+	if len(t0) != len(c.tables[0]) {
+		imin, imax = t0.getRange(c.s.icmp)
+	}
+	t1 = vt1.getOverlaps(t1, c.s.icmp, imin.ukey(), imax.ukey(), false)
+	// Get entire range covered by compaction.
+	amin, amax := append(t0, t1...).getRange(c.s.icmp)
 
 	// See if we can grow the number of inputs in "level" without
 	// changing the number of "level+1" files we pick up.
 	if len(t1) > 0 {
-		var exp0 tFiles
-		vt0.getOverlaps(amin.ukey(), amax.ukey(), &exp0, level != 0, s.icmp.ucmp)
-		if len(exp0) > len(t0) && t1.size()+exp0.size() < kExpCompactionMaxBytes {
-			var exp1 tFiles
-			xmin, xmax := exp0.getRange(s.icmp)
-			vt1.getOverlaps(xmin.ukey(), xmax.ukey(), &exp1, true, s.icmp.ucmp)
+		exp0 := vt0.getOverlaps(nil, c.s.icmp, amin.ukey(), amax.ukey(), c.level == 0)
+		if len(exp0) > len(t0) && t1.size()+exp0.size() < limit {
+			xmin, xmax := exp0.getRange(c.s.icmp)
+			exp1 := vt1.getOverlaps(nil, c.s.icmp, xmin.ukey(), xmax.ukey(), false)
 			if len(exp1) == len(t1) {
-				s.logf("table@compaction expanding L%d+L%d (F·%d S·%s)+(F·%d S·%s) -> (F·%d S·%s)+(F·%d S·%s)",
-					level, level+1, len(t0), shortenb(int(t0.size())), len(t1), shortenb(int(t1.size())),
+				c.s.logf("table@compaction expanding L%d+L%d (F·%d S·%s)+(F·%d S·%s) -> (F·%d S·%s)+(F·%d S·%s)",
+					c.level, c.level+1, len(t0), shortenb(int(t0.size())), len(t1), shortenb(int(t1.size())),
 					len(exp0), shortenb(int(exp0.size())), len(exp1), shortenb(int(exp1.size())))
-				min, max = xmin, xmax
+				imin, imax = xmin, xmax
 				t0, t1 = exp0, exp1
-				amin, amax = append(t0, t1...).getRange(s.icmp)
+				amin, amax = append(t0, t1...).getRange(c.s.icmp)
 			}
 		}
 	}
 
 	// Compute the set of grandparent files that overlap this compaction
 	// (parent == level+1; grandparent == level+2)
-	if level+2 < kNumLevels {
-		v.tables[level+2].getOverlaps(amin.ukey(), amax.ukey(), &c.gp, true, s.icmp.ucmp)
+	if c.level+2 < c.s.o.GetNumLevel() {
+		c.gp = c.v.tables[c.level+2].getOverlaps(c.gp, c.s.icmp, amin.ukey(), amax.ukey(), false)
 	}
 
 	c.tables[0], c.tables[1] = t0, t1
-	c.min, c.max = min, max
+	c.imin, c.imax = imin, imax
 }
 
 // Check whether compaction is trivial.
 func (c *compaction) trivial() bool {
-	return len(c.tables[0]) == 1 && len(c.tables[1]) == 0 && c.gp.size() <= kMaxGrandParentOverlapBytes
+	return len(c.tables[0]) == 1 && len(c.tables[1]) == 0 && c.gp.size() <= c.maxGPOverlaps
 }
 
-func (c *compaction) isBaseLevelForKey(key []byte) bool {
-	s := c.s
-	v := c.version
-
-	for level, tt := range v.tables[c.level+2:] {
-		for c.tPtrs[level] < len(tt) {
-			t := tt[c.tPtrs[level]]
-			if s.icmp.uCompare(key, t.max.ukey()) <= 0 {
-				// We've advanced far enough
-				if s.icmp.uCompare(key, t.min.ukey()) >= 0 {
-					// Key falls in this file's range, so definitely not base level
+func (c *compaction) baseLevelForKey(ukey []byte) bool {
+	for level, tables := range c.v.tables[c.level+2:] {
+		for c.tPtrs[level] < len(tables) {
+			t := tables[c.tPtrs[level]]
+			if c.s.icmp.uCompare(ukey, t.imax.ukey()) <= 0 {
+				// We've advanced far enough.
+				if c.s.icmp.uCompare(ukey, t.imin.ukey()) >= 0 {
+					// Key falls in this file's range, so definitely not base level.
 					return false
 				}
 				break
@@ -349,55 +398,61 @@ func (c *compaction) isBaseLevelForKey(key []byte) bool {
 	return true
 }
 
-func (c *compaction) shouldStopBefore(key iKey) bool {
-	for ; c.gpidx < len(c.gp); c.gpidx++ {
-		gp := c.gp[c.gpidx]
-		if c.s.icmp.Compare(key, gp.max) <= 0 {
+func (c *compaction) shouldStopBefore(ikey iKey) bool {
+	for ; c.gpi < len(c.gp); c.gpi++ {
+		gp := c.gp[c.gpi]
+		if c.s.icmp.Compare(ikey, gp.imax) <= 0 {
 			break
 		}
 		if c.seenKey {
-			c.overlappedBytes += gp.size
+			c.gpOverlappedBytes += gp.size
 		}
 	}
 	c.seenKey = true
 
-	if c.overlappedBytes > kMaxGrandParentOverlapBytes {
-		// Too much overlap for current output; start new output
-		c.overlappedBytes = 0
+	if c.gpOverlappedBytes > c.maxGPOverlaps {
+		// Too much overlap for current output; start new output.
+		c.gpOverlappedBytes = 0
 		return true
 	}
 	return false
 }
 
+// Creates an iterator.
 func (c *compaction) newIterator() iterator.Iterator {
-	s := c.s
-
-	level := c.level
-	icap := 2
+	// Creates iterator slice.
+	icap := len(c.tables)
 	if c.level == 0 {
+		// Special case for level-0
 		icap = len(c.tables[0]) + 1
 	}
 	its := make([]iterator.Iterator, 0, icap)
 
+	// Options.
 	ro := &opt.ReadOptions{
 		DontFillCache: true,
+		Strict:        opt.StrictOverride,
+	}
+	strict := c.s.o.GetStrict(opt.StrictCompaction)
+	if strict {
+		ro.Strict |= opt.StrictReader
 	}
-	strict := s.o.GetStrict(opt.StrictIterator)
 
-	for i, tt := range c.tables {
-		if len(tt) == 0 {
+	for i, tables := range c.tables {
+		if len(tables) == 0 {
 			continue
 		}
 
-		if level+i == 0 {
-			for _, t := range tt {
-				its = append(its, s.tops.newIterator(t, nil, ro))
+		// Level-0 is not sorted and may overlaps each other.
+		if c.level+i == 0 {
+			for _, t := range tables {
+				its = append(its, c.s.tops.newIterator(t, nil, ro))
 			}
 		} else {
-			it := iterator.NewIndexedIterator(tt.newIndexIterator(s.tops, s.icmp, nil, ro), strict, true)
+			it := iterator.NewIndexedIterator(tables.newIndexIterator(c.s.tops, c.s.icmp, nil, ro), strict)
 			its = append(its, it)
 		}
 	}
 
-	return iterator.NewMergedIterator(its, s.icmp, true)
+	return iterator.NewMergedIterator(its, c.s.icmp, strict)
 }

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/session_record.go

@@ -9,11 +9,11 @@ package leveldb
 import (
 	"bufio"
 	"encoding/binary"
-	"errors"
 	"io"
-)
+	"strings"
 
-var errCorruptManifest = errors.New("leveldb: corrupt manifest")
+	"github.com/syndtr/goleveldb/leveldb/errors"
+)
 
 type byteReader interface {
 	io.Reader
@@ -22,32 +22,28 @@ type byteReader interface {
 
 // These numbers are written to disk and should not be changed.
 const (
-	recComparer          = 1
-	recJournalNum        = 2
-	recNextNum           = 3
-	recSeq               = 4
-	recCompactionPointer = 5
-	recDeletedTable      = 6
-	recNewTable          = 7
+	recComparer    = 1
+	recJournalNum  = 2
+	recNextFileNum = 3
+	recSeqNum      = 4
+	recCompPtr     = 5
+	recDelTable    = 6
+	recAddTable    = 7
 	// 8 was used for large value refs
 	recPrevJournalNum = 9
 )
 
 type cpRecord struct {
 	level int
-	key   iKey
+	ikey  iKey
 }
 
-type ntRecord struct {
+type atRecord struct {
 	level int
 	num   uint64
 	size  uint64
-	min   iKey
-	max   iKey
-}
-
-func (r ntRecord) makeFile(s *session) *tFile {
-	return newTFile(s.getTableFile(r.num), r.size, r.min, r.max)
+	imin  iKey
+	imax  iKey
 }
 
 type dtRecord struct {
@@ -56,17 +52,20 @@ type dtRecord struct {
 }
 
 type sessionRecord struct {
-	hasRec             int
-	comparer           string
-	journalNum         uint64
-	prevJournalNum     uint64
-	nextNum            uint64
-	seq                uint64
-	compactionPointers []cpRecord
-	addedTables        []ntRecord
-	deletedTables      []dtRecord
-	scratch            [binary.MaxVarintLen64]byte
-	err                error
+	numLevel int
+
+	hasRec         int
+	comparer       string
+	journalNum     uint64
+	prevJournalNum uint64
+	nextFileNum    uint64
+	seqNum         uint64
+	compPtrs       []cpRecord
+	addedTables    []atRecord
+	deletedTables  []dtRecord
+
+	scratch [binary.MaxVarintLen64]byte
+	err     error
 }
 
 func (p *sessionRecord) has(rec int) bool {
@@ -88,47 +87,47 @@ func (p *sessionRecord) setPrevJournalNum(num uint64) {
 	p.prevJournalNum = num
 }
 
-func (p *sessionRecord) setNextNum(num uint64) {
-	p.hasRec |= 1 << recNextNum
-	p.nextNum = num
+func (p *sessionRecord) setNextFileNum(num uint64) {
+	p.hasRec |= 1 << recNextFileNum
+	p.nextFileNum = num
 }
 
-func (p *sessionRecord) setSeq(seq uint64) {
-	p.hasRec |= 1 << recSeq
-	p.seq = seq
+func (p *sessionRecord) setSeqNum(num uint64) {
+	p.hasRec |= 1 << recSeqNum
+	p.seqNum = num
 }
 
-func (p *sessionRecord) addCompactionPointer(level int, key iKey) {
-	p.hasRec |= 1 << recCompactionPointer
-	p.compactionPointers = append(p.compactionPointers, cpRecord{level, key})
+func (p *sessionRecord) addCompPtr(level int, ikey iKey) {
+	p.hasRec |= 1 << recCompPtr
+	p.compPtrs = append(p.compPtrs, cpRecord{level, ikey})
 }
 
-func (p *sessionRecord) resetCompactionPointers() {
-	p.hasRec &= ^(1 << recCompactionPointer)
-	p.compactionPointers = p.compactionPointers[:0]
+func (p *sessionRecord) resetCompPtrs() {
+	p.hasRec &= ^(1 << recCompPtr)
+	p.compPtrs = p.compPtrs[:0]
 }
 
-func (p *sessionRecord) addTable(level int, num, size uint64, min, max iKey) {
-	p.hasRec |= 1 << recNewTable
-	p.addedTables = append(p.addedTables, ntRecord{level, num, size, min, max})
+func (p *sessionRecord) addTable(level int, num, size uint64, imin, imax iKey) {
+	p.hasRec |= 1 << recAddTable
+	p.addedTables = append(p.addedTables, atRecord{level, num, size, imin, imax})
 }
 
 func (p *sessionRecord) addTableFile(level int, t *tFile) {
-	p.addTable(level, t.file.Num(), t.size, t.min, t.max)
+	p.addTable(level, t.file.Num(), t.size, t.imin, t.imax)
 }
 
 func (p *sessionRecord) resetAddedTables() {
-	p.hasRec &= ^(1 << recNewTable)
+	p.hasRec &= ^(1 << recAddTable)
 	p.addedTables = p.addedTables[:0]
 }
 
-func (p *sessionRecord) deleteTable(level int, num uint64) {
-	p.hasRec |= 1 << recDeletedTable
+func (p *sessionRecord) delTable(level int, num uint64) {
+	p.hasRec |= 1 << recDelTable
 	p.deletedTables = append(p.deletedTables, dtRecord{level, num})
 }
 
 func (p *sessionRecord) resetDeletedTables() {
-	p.hasRec &= ^(1 << recDeletedTable)
+	p.hasRec &= ^(1 << recDelTable)
 	p.deletedTables = p.deletedTables[:0]
 }
 
@@ -161,43 +160,45 @@ func (p *sessionRecord) encode(w io.Writer) error {
 		p.putUvarint(w, recJournalNum)
 		p.putUvarint(w, p.journalNum)
 	}
-	if p.has(recNextNum) {
-		p.putUvarint(w, recNextNum)
-		p.putUvarint(w, p.nextNum)
+	if p.has(recNextFileNum) {
+		p.putUvarint(w, recNextFileNum)
+		p.putUvarint(w, p.nextFileNum)
 	}
-	if p.has(recSeq) {
-		p.putUvarint(w, recSeq)
-		p.putUvarint(w, p.seq)
+	if p.has(recSeqNum) {
+		p.putUvarint(w, recSeqNum)
+		p.putUvarint(w, p.seqNum)
 	}
-	for _, cp := range p.compactionPointers {
-		p.putUvarint(w, recCompactionPointer)
-		p.putUvarint(w, uint64(cp.level))
-		p.putBytes(w, cp.key)
+	for _, r := range p.compPtrs {
+		p.putUvarint(w, recCompPtr)
+		p.putUvarint(w, uint64(r.level))
+		p.putBytes(w, r.ikey)
 	}
-	for _, t := range p.deletedTables {
-		p.putUvarint(w, recDeletedTable)
-		p.putUvarint(w, uint64(t.level))
-		p.putUvarint(w, t.num)
+	for _, r := range p.deletedTables {
+		p.putUvarint(w, recDelTable)
+		p.putUvarint(w, uint64(r.level))
+		p.putUvarint(w, r.num)
 	}
-	for _, t := range p.addedTables {
-		p.putUvarint(w, recNewTable)
-		p.putUvarint(w, uint64(t.level))
-		p.putUvarint(w, t.num)
-		p.putUvarint(w, t.size)
-		p.putBytes(w, t.min)
-		p.putBytes(w, t.max)
+	for _, r := range p.addedTables {
+		p.putUvarint(w, recAddTable)
+		p.putUvarint(w, uint64(r.level))
+		p.putUvarint(w, r.num)
+		p.putUvarint(w, r.size)
+		p.putBytes(w, r.imin)
+		p.putBytes(w, r.imax)
 	}
 	return p.err
 }
 
-func (p *sessionRecord) readUvarint(r io.ByteReader) uint64 {
+func (p *sessionRecord) readUvarintMayEOF(field string, r io.ByteReader, mayEOF bool) uint64 {
 	if p.err != nil {
 		return 0
 	}
 	x, err := binary.ReadUvarint(r)
 	if err != nil {
-		if err == io.EOF {
-			p.err = errCorruptManifest
+		if err == io.ErrUnexpectedEOF || (mayEOF == false && err == io.EOF) {
+			p.err = errors.NewErrCorrupted(nil, &ErrManifestCorrupted{field, "short read"})
+		} else if strings.HasPrefix(err.Error(), "binary:") {
+			p.err = errors.NewErrCorrupted(nil, &ErrManifestCorrupted{field, err.Error()})
 		} else {
 			p.err = err
 		}
@@ -206,35 +207,39 @@ func (p *sessionRecord) readUvarint(r io.ByteReader) uint64 {
 	return x
 }
 
-func (p *sessionRecord) readBytes(r byteReader) []byte {
+func (p *sessionRecord) readUvarint(field string, r io.ByteReader) uint64 {
+	return p.readUvarintMayEOF(field, r, false)
+}
+
+func (p *sessionRecord) readBytes(field string, r byteReader) []byte {
 	if p.err != nil {
 		return nil
 	}
-	n := p.readUvarint(r)
+	n := p.readUvarint(field, r)
 	if p.err != nil {
 		return nil
 	}
 	x := make([]byte, n)
 	_, p.err = io.ReadFull(r, x)
 	if p.err != nil {
-		if p.err == io.EOF {
-			p.err = errCorruptManifest
+		if p.err == io.ErrUnexpectedEOF {
+			p.err = errors.NewErrCorrupted(nil, &ErrManifestCorrupted{field, "short read"})
 		}
 		return nil
 	}
 	return x
 }
 
-func (p *sessionRecord) readLevel(r io.ByteReader) int {
+func (p *sessionRecord) readLevel(field string, r io.ByteReader) int {
 	if p.err != nil {
 		return 0
 	}
-	x := p.readUvarint(r)
+	x := p.readUvarint(field, r)
 	if p.err != nil {
 		return 0
 	}
-	if x >= kNumLevels {
-		p.err = errCorruptManifest
+	if x >= uint64(p.numLevel) {
+		p.err = errors.NewErrCorrupted(nil, &ErrManifestCorrupted{field, "invalid level number"})
 		return 0
 	}
 	return int(x)
@@ -247,59 +252,59 @@ func (p *sessionRecord) decode(r io.Reader) error {
 	}
 	p.err = nil
 	for p.err == nil {
-		rec, err := binary.ReadUvarint(br)
-		if err != nil {
-			if err == io.EOF {
-				err = nil
+		rec := p.readUvarintMayEOF("field-header", br, true)
+		if p.err != nil {
+			if p.err == io.EOF {
+				return nil
 			}
-			return err
+			return p.err
 		}
 		switch rec {
 		case recComparer:
-			x := p.readBytes(br)
+			x := p.readBytes("comparer", br)
 			if p.err == nil {
 				p.setComparer(string(x))
 			}
 		case recJournalNum:
-			x := p.readUvarint(br)
+			x := p.readUvarint("journal-num", br)
 			if p.err == nil {
 				p.setJournalNum(x)
 			}
 		case recPrevJournalNum:
-			x := p.readUvarint(br)
+			x := p.readUvarint("prev-journal-num", br)
 			if p.err == nil {
 				p.setPrevJournalNum(x)
 			}
-		case recNextNum:
-			x := p.readUvarint(br)
+		case recNextFileNum:
+			x := p.readUvarint("next-file-num", br)
 			if p.err == nil {
-				p.setNextNum(x)
+				p.setNextFileNum(x)
 			}
-		case recSeq:
-			x := p.readUvarint(br)
+		case recSeqNum:
+			x := p.readUvarint("seq-num", br)
 			if p.err == nil {
-				p.setSeq(x)
+				p.setSeqNum(x)
 			}
-		case recCompactionPointer:
-			level := p.readLevel(br)
-			key := p.readBytes(br)
+		case recCompPtr:
+			level := p.readLevel("comp-ptr.level", br)
+			ikey := p.readBytes("comp-ptr.ikey", br)
 			if p.err == nil {
-				p.addCompactionPointer(level, iKey(key))
+				p.addCompPtr(level, iKey(ikey))
 			}
-		case recNewTable:
-			level := p.readLevel(br)
-			num := p.readUvarint(br)
-			size := p.readUvarint(br)
-			min := p.readBytes(br)
-			max := p.readBytes(br)
+		case recAddTable:
+			level := p.readLevel("add-table.level", br)
+			num := p.readUvarint("add-table.num", br)
+			size := p.readUvarint("add-table.size", br)
+			imin := p.readBytes("add-table.imin", br)
+			imax := p.readBytes("add-table.imax", br)
 			if p.err == nil {
-				p.addTable(level, num, size, min, max)
+				p.addTable(level, num, size, imin, imax)
 			}
-		case recDeletedTable:
-			level := p.readLevel(br)
-			num := p.readUvarint(br)
+		case recDelTable:
+			level := p.readLevel("del-table.level", br)
+			num := p.readUvarint("del-table.num", br)
 			if p.err == nil {
-				p.deleteTable(level, num)
+				p.delTable(level, num)
 			}
 		}
 	}

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/session_record_test.go

@@ -9,6 +9,8 @@ package leveldb
 import (
 	"bytes"
 	"testing"
+
+	"github.com/syndtr/goleveldb/leveldb/opt"
 )
 
 func decodeEncode(v *sessionRecord) (res bool, err error) {
@@ -17,7 +19,7 @@ func decodeEncode(v *sessionRecord) (res bool, err error) {
 	if err != nil {
 		return
 	}
-	v2 := new(sessionRecord)
+	v2 := &sessionRecord{numLevel: opt.DefaultNumLevel}
 	err = v.decode(b)
 	if err != nil {
 		return
@@ -32,7 +34,7 @@ func decodeEncode(v *sessionRecord) (res bool, err error) {
 
 func TestSessionRecord_EncodeDecode(t *testing.T) {
 	big := uint64(1) << 50
-	v := new(sessionRecord)
+	v := &sessionRecord{numLevel: opt.DefaultNumLevel}
 	i := uint64(0)
 	test := func() {
 		res, err := decodeEncode(v)
@@ -47,16 +49,16 @@ func TestSessionRecord_EncodeDecode(t *testing.T) {
 	for ; i < 4; i++ {
 		test()
 		v.addTable(3, big+300+i, big+400+i,
-			newIKey([]byte("foo"), big+500+1, tVal),
-			newIKey([]byte("zoo"), big+600+1, tDel))
-		v.deleteTable(4, big+700+i)
-		v.addCompactionPointer(int(i), newIKey([]byte("x"), big+900+1, tVal))
+			newIkey([]byte("foo"), big+500+1, ktVal),
+			newIkey([]byte("zoo"), big+600+1, ktDel))
+		v.delTable(4, big+700+i)
+		v.addCompPtr(int(i), newIkey([]byte("x"), big+900+1, ktVal))
 	}
 
 	v.setComparer("foo")
 	v.setJournalNum(big + 100)
 	v.setPrevJournalNum(big + 99)
-	v.setNextNum(big + 200)
-	v.setSeq(big + 1000)
+	v.setNextFileNum(big + 200)
+	v.setSeqNum(big + 1000)
 	test()
 }

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/session_util.go

@@ -14,7 +14,7 @@ import (
 	"github.com/syndtr/goleveldb/leveldb/storage"
 )
 
-// logging
+// Logging.
 
 type dropper struct {
 	s    *session
@@ -22,22 +22,17 @@ type dropper struct {
 }
 
 func (d dropper) Drop(err error) {
-	if e, ok := err.(journal.DroppedError); ok {
+	if e, ok := err.(*journal.ErrCorrupted); ok {
 		d.s.logf("journal@drop %s-%d S·%s %q", d.file.Type(), d.file.Num(), shortenb(e.Size), e.Reason)
 	} else {
 		d.s.logf("journal@drop %s-%d %q", d.file.Type(), d.file.Num(), err)
 	}
 }
 
-func (s *session) log(v ...interface{}) {
-	s.stor.Log(fmt.Sprint(v...))
-}
+func (s *session) log(v ...interface{})                 { s.stor.Log(fmt.Sprint(v...)) }
+func (s *session) logf(format string, v ...interface{}) { s.stor.Log(fmt.Sprintf(format, v...)) }
 
-func (s *session) logf(format string, v ...interface{}) {
-	s.stor.Log(fmt.Sprintf(format, v...))
-}
-
-// file utils
+// File utils.
 
 func (s *session) getJournalFile(num uint64) storage.File {
 	return s.stor.GetFile(num, storage.TypeJournal)
@@ -56,9 +51,14 @@ func (s *session) newTemp() storage.File {
 	return s.stor.GetFile(num, storage.TypeTemp)
 }
 
-// session state
+func (s *session) tableFileFromRecord(r atRecord) *tFile {
+	return newTableFile(s.getTableFile(r.num), r.size, r.imin, r.imax)
+}
 
-// Get current version.
+// Session state.
+
+// Get current version. This will incr version ref, must call
+// version.release (exactly once) after use.
 func (s *session) version() *version {
 	s.vmu.Lock()
 	defer s.vmu.Unlock()
@@ -66,85 +66,80 @@ func (s *session) version() *version {
 	return s.stVersion
 }
 
-// Get current version; no barrier.
-func (s *session) version_NB() *version {
-	return s.stVersion
-}
-
 // Set current version to v.
 func (s *session) setVersion(v *version) {
 	s.vmu.Lock()
-	v.ref = 1
+	v.ref = 1 // Holds by session.
 	if old := s.stVersion; old != nil {
-		v.ref++
+		v.ref++ // Holds by old version.
 		old.next = v
-		old.release_NB()
+		old.releaseNB()
 	}
 	s.stVersion = v
 	s.vmu.Unlock()
 }
 
 // Get current unused file number.
-func (s *session) fileNum() uint64 {
-	return atomic.LoadUint64(&s.stFileNum)
+func (s *session) nextFileNum() uint64 {
+	return atomic.LoadUint64(&s.stNextFileNum)
 }
 
-// Get current unused file number to num.
-func (s *session) setFileNum(num uint64) {
-	atomic.StoreUint64(&s.stFileNum, num)
+// Set current unused file number to num.
+func (s *session) setNextFileNum(num uint64) {
+	atomic.StoreUint64(&s.stNextFileNum, num)
 }
 
 // Mark file number as used.
 func (s *session) markFileNum(num uint64) {
-	num += 1
+	nextFileNum := num + 1
 	for {
-		old, x := s.stFileNum, num
+		old, x := s.stNextFileNum, nextFileNum
 		if old > x {
 			x = old
 		}
-		if atomic.CompareAndSwapUint64(&s.stFileNum, old, x) {
+		if atomic.CompareAndSwapUint64(&s.stNextFileNum, old, x) {
 			break
 		}
 	}
 }
 
 // Allocate a file number.
-func (s *session) allocFileNum() (num uint64) {
-	return atomic.AddUint64(&s.stFileNum, 1) - 1
+func (s *session) allocFileNum() uint64 {
+	return atomic.AddUint64(&s.stNextFileNum, 1) - 1
 }
 
 // Reuse given file number.
 func (s *session) reuseFileNum(num uint64) {
 	for {
-		old, x := s.stFileNum, num
+		old, x := s.stNextFileNum, num
 		if old != x+1 {
 			x = old
 		}
-		if atomic.CompareAndSwapUint64(&s.stFileNum, old, x) {
+		if atomic.CompareAndSwapUint64(&s.stNextFileNum, old, x) {
 			break
 		}
 	}
 }
 
-// manifest related utils
+// Manifest related utils.
 
 // Fill given session record obj with current states; need external
 // synchronization.
 func (s *session) fillRecord(r *sessionRecord, snapshot bool) {
-	r.setNextNum(s.fileNum())
+	r.setNextFileNum(s.nextFileNum())
 
 	if snapshot {
 		if !r.has(recJournalNum) {
 			r.setJournalNum(s.stJournalNum)
 		}
 
-		if !r.has(recSeq) {
-			r.setSeq(s.stSeq)
+		if !r.has(recSeqNum) {
+			r.setSeqNum(s.stSeqNum)
 		}
 
-		for level, ik := range s.stCPtrs {
+		for level, ik := range s.stCompPtrs {
 			if ik != nil {
-				r.addCompactionPointer(level, ik)
+				r.addCompPtr(level, ik)
 			}
 		}
 
@@ -152,7 +147,7 @@ func (s *session) fillRecord(r *sessionRecord, snapshot bool) {
 	}
 }
 
-// Mark if record has been commited, this will update session state;
+// Mark if record has been committed, this will update session state;
 // need external synchronization.
 func (s *session) recordCommited(r *sessionRecord) {
 	if r.has(recJournalNum) {
@@ -163,12 +158,12 @@ func (s *session) recordCommited(r *sessionRecord) {
 		s.stPrevJournalNum = r.prevJournalNum
 	}
 
-	if r.has(recSeq) {
-		s.stSeq = r.seq
+	if r.has(recSeqNum) {
+		s.stSeqNum = r.seqNum
 	}
 
-	for _, p := range r.compactionPointers {
-		s.stCPtrs[p.level] = iKey(p.key)
+	for _, p := range r.compPtrs {
+		s.stCompPtrs[p.level] = iKey(p.ikey)
 	}
 }
 
@@ -183,10 +178,11 @@ func (s *session) newManifest(rec *sessionRecord, v *version) (err error) {
 	jw := journal.NewWriter(writer)
 
 	if v == nil {
-		v = s.version_NB()
+		v = s.version()
+		defer v.release()
 	}
 	if rec == nil {
-		rec = new(sessionRecord)
+		rec = &sessionRecord{numLevel: s.o.GetNumLevel()}
 	}
 	s.fillRecord(rec, true)
 	v.fillRecord(rec)

Godeps/_workspace/src/github.com/syndtr/goleveldb/leveldb/storage/file_storage.go

@@ -344,19 +344,17 @@ type fileWrap struct {
 }
 
 func (fw fileWrap) Sync() error {
+	if err := fw.File.Sync(); err != nil {
+		return err
+	}
 	if fw.f.Type() == TypeManifest {
 		// Also sync parent directory if file type is manifest.
 		// See: https://code.google.com/p/leveldb/issues/detail?id=190.
-		f, err := os.Open(fw.f.fs.path)
-		if err != nil {
-			return err
-		}
-		defer f.Close()
-		if err := f.Sync(); err != nil {
+		if err := syncDir(fw.f.fs.path); err != nil {
 			return err
 		}
 	}
-	return fw.File.Sync()
+	return nil
 }
 
 func (fw fileWrap) Close() error {